Skip to content

Instantly share code, notes, and snippets.

@HDCharles
Last active July 8, 2026 03:03
Show Gist options
  • Select an option

  • Save HDCharles/edec551a407956861815b1f87ef7e52c to your computer and use it in GitHub Desktop.

Select an option

Save HDCharles/edec551a407956861815b1f87ef7e52c to your computer and use it in GitHub Desktop.
download_analysis
Command: python hf_download_model.py
Loading data from hf_quant_dataset.csv (use --retry for failed rows, --fetch-all to re-fetch everything)
Dataset: 452 quantized models with known base date and model size
Hubs: {'RedHatAI': 379, 'nvidia': 73}
Formats: {'fp8': 180, 'w4a16': 75, 'w8a8': 74, 'nvfp4': 65, 'w8a16': 33, 'deepsparse': 8, 'int4': 6, 'gguf': 6, 'unknown': 2, 'gptq': 1, 'fp4-qad': 1, 'fp4-eagle3': 1}
Prior variant exists: {1: 237, 0: 215}
======================================================================
Model 1: All models — hub, format, size, base popularity
R² = 0.4129 (n = 452)
======================================================================
Feature Coefficient Interpretation
--------------------------------------------------------------------------------
(intercept) 2.6371 baseline log-downloads
quant_format_int4 -9.1491 -100.0% vs baseline category
hub_nvidia 2.2113 +812.8% vs baseline category
quant_format_w8a16 -1.9376 -85.6% vs baseline category
quant_format_nvfp4 1.8961 +566.0% vs baseline category
quant_format_gguf -1.8140 -83.7% vs baseline category
quant_format_fp8 0.6878 +98.9% vs baseline category
log_base_downloads 0.4158 +51.56% downloads per unit increase
quant_format_w4a16 0.3209 +37.8% vs baseline category
quant_format_w8a8 -0.1884 -17.2% vs baseline category
quant_format_infrequent_sklearn -0.0837 -8.0% vs baseline category
days_since_base -0.0092 -0.92% downloads per unit increase
model_size_b -0.0010 -0.10% downloads per unit increase
days_since_release 0.0006 +0.06% downloads per unit increase
======================================================================
Top 10 over-performers (actual >> predicted)
======================================================================
hub name downloads predicted_dl quant_format days_since_base
RedHatAI Llama-2-7b-chat-quantized.w8a8 14378 11 w8a8 368
RedHatAI Llama-2-7b-chat-quantized.w8a16 2589 1 w8a16 359
RedHatAI Qwen2.5-1.5B-quantized.w8a8 6618774 12408 w8a8 24
nvidia NVIDIA-Nemotron-3-Nano-4B-GGUF 69030 201 gguf 308
RedHatAI Mixtral-8x7B-Instruct-v0.1-FP8 154925 663 fp8 452
nvidia NVIDIA-Nemotron-3-Super-120B-A12B-FP8 2232273 10318 fp8 4
RedHatAI Llama-Guard-4-12B-quantized.w4a16 83462 441 w4a16 299
nvidia NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 5547346 34546 nvfp4 4
RedHatAI Llama-3.2-1B-Instruct-FP8-dynamic 11682743 77108 fp8 7
RedHatAI gemma-4-31B-it-FP8-block 4358837 31112 fp8 23
======================================================================
Top 10 under-performers (actual << predicted)
======================================================================
hub name downloads predicted_dl quant_format days_since_base
nvidia Wan2.2-T2V-A14B-Diffusers-NVFP4 0 22844 nvfp4 289
nvidia Wan2.2-T2V-A14B-Diffusers-FP8 0 6823 fp8 289
RedHatAI Qwen2.5-Math-7B-FP8-dynamic 16 10811 fp8 73
RedHatAI Qwen2.5-32B-FP8-dynamic 18 11141 fp8 74
RedHatAI Qwen2.5-32B-quantized.w4a16 21 11566 w4a16 33
RedHatAI QwQ-32B-Preview-FP8-dynamic 26 9848 fp8 37
RedHatAI Qwen2.5-1.5B-Instruct-FP8-dynamic 167 58366 fp8 71
RedHatAI Qwen2.5-32B-quantized.w8a8 22 7595 w8a8 24
RedHatAI Qwen2.5-72B-quantized.w4a16 18 5219 w4a16 32
RedHatAI Qwen3-VL-32B-Instruct-FP8-block 108 17619 fp8 52
======================================================================
Mean downloads by quant format
======================================================================
count mean median
quant_format
gptq 1 1388401 1388401
nvfp4 65 852222 91426
fp8 180 364859 17039
w8a8 74 134433 5544
w4a16 75 104454 8164
unknown 2 98039 98039
gguf 6 11620 136
w8a16 33 11413 776
fp4-qad 1 4834 4834
int4 6 338 0
deepsparse 8 291 281
fp4-eagle3 1 260 260
======================================================================
Model 2: RedHatAI only — format × period interaction
R² = 0.4066 (n = 379)
======================================================================
Feature Coef Multiplier Interpretation
----------------------------------------------------------------------------------------------------
(intercept) 0.414 baseline
format_x_period_w4a16:2026H1 3.885 48.66 ×48.66 vs baseline
format_x_period_fp8:2024H1 3.842 46.62 ×46.62 vs baseline
format_x_period_nvfp4:2026H1 3.818 45.52 ×45.52 vs baseline
format_x_period_nvfp4:2025H2 3.735 41.88 ×41.88 vs baseline
format_x_period_nvfp4:2025H1 3.187 24.23 ×24.23 vs baseline
format_x_period_fp8:2026H1 3.097 22.14 ×22.14 vs baseline
format_x_period_fp8:2025H1 3.003 20.14 ×20.14 vs baseline
format_x_period_w8a8:2025H1 2.920 18.55 ×18.55 vs baseline
format_x_period_w4a16:2025H1 2.865 17.54 ×17.54 vs baseline
format_x_period_fp8:2024H2 2.524 12.48 ×12.48 vs baseline
format_x_period_infrequent_sklearn 2.202 9.04 ×9.04 vs baseline
format_x_period_w4a16:2024H2 2.021 7.55 ×7.55 vs baseline
format_x_period_w4a16:2025H2 1.681 5.37 ×5.37 vs baseline
format_x_period_w8a8:2024H2 1.581 4.86 ×4.86 vs baseline
format_x_period_fp8:2025H2 1.385 3.99 ×3.99 vs baseline
format_x_period_gguf:2024H2 -1.296 0.27 ×0.27 vs baseline
within_1_month 1.185 3.27 ×3.27 when released within 1 month of base
format_x_period_w8a8:2026H1 1.121 3.07 ×3.07 vs baseline
within_1_day 1.055 2.87 ×2.87 when released within 1 day of base
within_1_week 0.861 2.37 ×2.37 when released within 1 week of base
log_base_downloads 0.380 1.46 +46.22% per unit
format_x_period_w8a16:2024H2 0.177 1.19 ×1.19 vs baseline
days_since_base -0.002 1.00 -0.21% per unit
model_size_b -0.002 1.00 -0.18% per unit
======================================================================
Format interest over time (coefficient = log-multiplier on downloads)
Positive = more downloads than format average, negative = fewer
======================================================================
Format 2024H1 2024H2 2025H1 2025H2 2026H1
-----------------------------------------------------------------
fp8 3.84 2.52 3.00 1.38 3.10
gguf — -1.30 — — —
nvfp4 — — 3.19 3.73 3.82
w4a16 — 2.02 2.86 1.68 3.88
w8a16 — 0.18 — — —
w8a8 — 1.58 2.92 — 1.12
======================================================================
Same table as download multiplier (1.00 = neutral)
======================================================================
Format 2024H1 2024H2 2025H1 2025H2 2026H1
-----------------------------------------------------------------
fp8 46.62 12.48 20.14 3.99 22.14
gguf — 0.27 — — —
nvfp4 — — 24.23 41.88 45.52
w4a16 — 7.55 17.54 5.37 48.66
w8a16 — 1.19 — — —
w8a8 — 4.86 18.55 — 3.07
======================================================================
Sample counts per format × period
======================================================================
year_half 2024H1 2024H2 2025H1 2025H2 2026H1
quant_format
deepsparse 8 0 0 0 0
fp8 12 52 37 29 23
gguf 0 5 0 0 0
gptq 1 0 0 0 0
nvfp4 0 0 5 14 11
w4a16 0 24 36 11 4
w8a16 0 33 0 0 0
w8a8 0 37 26 2 9
======================================================================
Model 3: RedHatAI only — prior variant exists (no interactions, no timing)
R² = 0.3349 (n = 379)
======================================================================
Feature Coef Multiplier Interpretation
----------------------------------------------------------------------------------------------------
(intercept) 3.429 baseline
quant_format_infrequent_sklearn 3.373 29.17 ×29.17 vs baseline
quant_format_gguf -2.762 0.06 ×0.06 vs baseline
quant_format_w8a16 -1.888 0.15 ×0.15 vs baseline
quant_format_nvfp4 1.630 5.11 ×5.11 vs baseline
quant_format_fp8 0.649 1.91 ×1.91 vs baseline
quant_format_w4a16 0.346 1.41 ×1.41 vs baseline
log_base_downloads 0.327 1.39 +38.73% per unit
total_variants_count 0.277 1.32 +31.93% per additional variant (RH+nvidia)
prior_variant_exists -0.238 0.79 ×0.79 when any prior variant exists
prior_same_format_exists 0.083 1.09 ×1.09 when same format already released
quant_format_w8a8 -0.071 0.93 ×0.93 vs baseline
days_since_base -0.008 0.99 -0.82% per unit
model_size_b -0.003 1.00 -0.26% per unit
days_since_prior_variant -0.001 1.00 -0.09% per day since prior variant published
======================================================================
Downloads by first-mover status (RedHatAI)
======================================================================
First quant variant for base: n=182, mean=293,267, median=8,064
Prior variant existed: n=197, mean=138,678, median=7,881
First of this quant format for base: n=345, mean=208,504, median=7,422
Same format already existed: n=34, mean=257,657, median=23,134
======================================================================
Model 4: RedHatAI only — format × size + prior variant exists + timing
R² = 0.4010 (n = 379)
======================================================================
Feature Coef Multiplier Interpretation
----------------------------------------------------------------------------------------------------
(intercept) -0.113 baseline
format_x_size_nvfp4:15-50B 3.411 30.29 ×30.29 vs baseline
format_x_size_nvfp4:5-15B 3.216 24.93 ×24.93 vs baseline
format_x_size_w4a16:5-15B 2.743 15.53 ×15.53 vs baseline
format_x_size_fp8:15-50B 2.677 14.54 ×14.54 vs baseline
format_x_size_w8a8:5-15B 2.627 13.84 ×13.84 vs baseline
format_x_size_fp8:5-15B 2.591 13.35 ×13.35 vs baseline
format_x_size_fp8:<5B 2.418 11.23 ×11.23 vs baseline
format_x_size_w4a16:15-50B 2.391 10.93 ×10.93 vs baseline
format_x_size_w8a8:<5B 2.167 8.73 ×8.73 vs baseline
format_x_size_w8a16:5-15B 1.916 6.79 ×6.79 vs baseline
format_x_size_w4a16:<5B 1.839 6.29 ×6.29 vs baseline
within_1_day 1.638 5.14 ×5.14 when released within 1 day of base
format_x_size_w8a16:<5B -1.392 0.25 ×0.25 vs baseline
within_1_month 1.377 3.96 ×3.96 when released within 1 month of base
format_x_size_w8a8:15-50B 1.139 3.13 ×3.13 vs baseline
format_x_size_gguf:<5B -0.658 0.52 ×0.52 vs baseline
within_1_week 0.570 1.77 ×1.77 when released within 1 week of base
log_base_downloads 0.372 1.45 +45.04% per unit
format_x_size_w8a16:15-50B -0.302 0.74 ×0.74 vs baseline
prior_variant_exists 0.210 1.23 ×1.23 when any prior variant exists
total_variants_count 0.153 1.17 +16.52% per additional variant (RH+nvidia)
format_x_size_infrequent_sklearn -0.062 0.94 ×0.94 vs baseline
prior_same_format_exists 0.040 1.04 ×1.04 when same format already released
days_since_base -0.003 1.00 -0.31% per unit
days_since_prior_variant -0.002 1.00 -0.15% per day since prior variant published
======================================================================
Format effectiveness by model size (coefficient = log-multiplier on downloads)
Positive = more downloads than format average, negative = fewer
======================================================================
Format <5B 5-15B 15-50B
---------------------------------------------
fp8 2.42 2.59 2.68
gguf -0.66 — —
nvfp4 — 3.22 3.41
w4a16 1.84 2.74 2.39
w8a16 -1.39 1.92 -0.30
w8a8 2.17 2.63 1.14
======================================================================
Same table as download multiplier (1.00 = neutral)
======================================================================
Format <5B 5-15B 15-50B
---------------------------------------------
fp8 11.23 13.35 14.54
gguf 0.52 — —
nvfp4 — 24.93 30.29
w4a16 6.29 15.53 10.93
w8a16 0.25 6.79 0.74
w8a8 8.73 13.84 3.13
======================================================================
Sample counts per format × size
======================================================================
size_bucket 15-50B 5-15B <5B
quant_format
deepsparse 0 8 0
fp8 73 50 30
gguf 0 2 3
gptq 0 1 0
nvfp4 24 6 0
w4a16 29 25 21
w8a16 11 10 12
w8a8 26 22 26
======================================================================
Downloads by first-mover status (RedHatAI)
======================================================================
First quant variant for base: n=182, mean=293,267, median=8,064
Prior variant existed: n=197, mean=138,678, median=7,881
First of this quant format for base: n=345, mean=208,504, median=7,422
Same format already existed: n=34, mean=257,657, median=23,134
======================================================================
Downloads by release timing relative to base model (RedHatAI)
======================================================================
Within 1 day (0-1d): n=11, mean=207,404, median=324
Within 1 week (2-7d): n=48, mean=683,608, median=37,746
Within 1 month (8-30d): n=176, mean=411,720, median=23,498
After 1 month (31+d): n=203, mean=40,549, median=3,624
#!/usr/bin/env python3
"""Train a linear model to predict quantized-model downloads from hub, time-to-release, format, and size."""
import argparse
import json
import os
import re
import sys
import urllib.parse
import urllib.request
from datetime import datetime
import numpy as np
import pandas as pd
from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import OneHotEncoder
CSV_PATH = "hf_quant_dataset.csv"
# ---------------------------------------------------------------------------
# Data fetching (shared logic with hf_quant_compare.py)
# ---------------------------------------------------------------------------
QUANT_KEYWORDS = [
"FP4", "FP8", "GPTQ", "AWQ", "INT4", "INT8", "NVFP4",
"W4A16", "W8A8", "W8A16", "W4A4", "quantized", "GGUF", "AutoFP8",
]
ORG_PREFIXES = [
("qwen", "Qwen"),
("llama", "meta-llama"),
("gemma", "google"),
("diffusiongem","google"),
("deepseek", "deepseek-ai"),
("glm", "THUDM"),
("mistral", "mistralai"),
("mixtral", "mistralai"),
("devstral", "mistralai"),
("phi-", "microsoft"),
("kimi", "moonshotai"),
("nemotron", "nvidia"),
("nvidia-nemotron", "nvidia"),
("minimax", "MiniMaxAI"),
("granite", "ibm-granite"),
("smollm", "HuggingFaceTB"),
("command", "CohereForAI"),
("whisper", "openai"),
("apertus", "ibm-granite"),
("omnicoder", "microsoft"),
("trinity", "Snowflake"),
]
def fetch_models(author):
url = (
f"https://huggingface.co/api/models?author={author}&limit=10000"
"&expand%5B%5D=downloadsAllTime"
"&expand%5B%5D=downloads"
"&expand%5B%5D=createdAt"
)
with urllib.request.urlopen(url) as resp:
return json.loads(resp.read())
def is_quantized(name):
return any(kw.lower() in name.lower() for kw in QUANT_KEYWORDS)
SUFFIXES_RE = re.compile(
r"[-_]("
r"NVFP4A16|NVFP4-FP8-BLOCK|NVFP4-FP8|NVFP4-v\d+|NVFP4-QAD|NVFP4"
r"|FP8-dynamic|FP8-Dynamic|FP8-BLOCK|FP8-block|FP8-Dflash|FP8-KV|FP8"
r"|FP4-QAD|FP4-Eagle\d+"
r"|GPTQ-\d+bit|GPTQ|AutoFP8"
r"|quantized\.w\d+a\d+|quantized-deepsparse"
r"|W4A16-G\d+|W4A16"
r"|GGUF"
r"|ONNX-INT4"
r")$",
re.IGNORECASE,
)
FIRST_QUANT_RE = re.compile(
r"[-_.]("
r"NVFP4A16|NVFP4|FP8|FP4|GPTQ|AutoFP8|quantized|W4A16|W8A8|W8A16|W4A4|GGUF|ONNX-INT4"
r")(?:[-_.\b]|$)",
re.IGNORECASE,
)
def extract_base(name):
m = FIRST_QUANT_RE.search(name)
if m:
return name[:m.start()]
return name
def normalize_base(base):
b = base.lower()
b = re.sub(r"^meta-", "", b)
b = b.replace("_", "-")
b = re.sub(r"-it$", "", b)
return b
def extract_quant_format(name):
"""Return the quantization format tag from a model name."""
m = SUFFIXES_RE.search(name)
if not m:
return "unknown"
fmt = m.group(1).lower()
if fmt.startswith("quantized."):
return fmt.replace("quantized.", "") # e.g. "w4a16"
if fmt.startswith("quantized"):
return fmt.replace("quantized-", "")
if "nvfp4" in fmt:
return "nvfp4"
if "fp8" in fmt:
return "fp8"
if "gptq" in fmt:
return "gptq"
if "autofp8" in fmt:
return "fp8"
if "gguf" in fmt:
return "gguf"
if "onnx-int4" in fmt:
return "int4"
return fmt
SIZE_RE = re.compile(r"(\d+(?:\.\d+)?)\s*[Bb]\b")
def extract_model_size_b(name):
"""Pull the largest 'NB' parameter count (in billions) from name. Returns float or NaN."""
matches = SIZE_RE.findall(name)
if not matches:
return float("nan")
return max(float(x) for x in matches)
# ---------------------------------------------------------------------------
# Original-date resolution (with caching)
# ---------------------------------------------------------------------------
def _fetch_date(model_id):
try:
url = f"https://huggingface.co/api/models/{model_id}"
with urllib.request.urlopen(url, timeout=10) as resp:
data = json.loads(resp.read())
c = data.get("createdAt", "")
return c[:10] if c else None
except Exception:
return None
def _fetch_downloads(model_id):
try:
url = f"https://huggingface.co/api/models/{model_id}?expand%5B%5D=downloadsAllTime"
with urllib.request.urlopen(url, timeout=10) as resp:
data = json.loads(resp.read())
return data.get("downloadsAllTime", 0)
except Exception:
return None
def _search_date(base_name):
try:
url = f"https://huggingface.co/api/models?search={urllib.parse.quote(base_name)}&limit=20"
with urllib.request.urlopen(url, timeout=10) as resp:
models = json.loads(resp.read())
best_date, best_id = None, None
for m in models:
n = m["id"].split("/")[-1]
if is_quantized(n):
continue
c = (m.get("createdAt") or "")[:10]
if c and (best_date is None or c < best_date):
best_date = c
best_id = m["id"]
return best_date, best_id
except Exception:
return None, None
def _guess_candidates(norm_base, sample_name):
base_cased = extract_base(sample_name)
base_no_meta = re.sub(r"^Meta-", "", base_cased)
for prefix, org in ORG_PREFIXES:
if norm_base.startswith(prefix):
cands = [f"{org}/{base_cased}", f"{org}/{base_no_meta}"]
if norm_base.startswith("gemma") or norm_base.startswith("diffusiongem"):
cands += [f"{org}/{base_cased}-it", f"{org}/{base_no_meta}-it"]
return cands
return []
def resolve_original_info(norm_base, sample_name, cache):
"""Look up the original model's release date and downloads. Returns (date, downloads)."""
if norm_base in cache:
return cache[norm_base]
for cand in _guess_candidates(norm_base, sample_name):
date = _fetch_date(cand)
if date:
dl = _fetch_downloads(cand)
cache[norm_base] = (date, dl)
return (date, dl)
date, repo_id = _search_date(extract_base(sample_name))
dl = _fetch_downloads(repo_id) if repo_id else None
cache[norm_base] = (date, dl)
return (date, dl)
# ---------------------------------------------------------------------------
# Build dataset
# ---------------------------------------------------------------------------
def _resolve_bases(df):
"""Resolve base-model dates and downloads, marking failures."""
cache = {}
bases = df["norm_base"].unique()
failed = []
print(f"\nResolving original release dates for {len(bases)} base models...")
for i, base in enumerate(sorted(bases), 1):
sample = df[df["norm_base"] == base].iloc[0]["name"]
resolve_original_info(base, sample, cache)
date, dl = cache.get(base, (None, None))
if date is None:
failed.append(base)
sys.stdout.write(f"\r [{i}/{len(bases)}]")
sys.stdout.flush()
print()
if failed:
print(f" WARNING: {len(failed)} base models failed to resolve:")
for b in failed:
print(f" - {b}")
return cache, failed
def _add_derived_columns(df):
"""Add all derived columns needed by models."""
df["log_base_downloads"] = np.log1p(df["base_downloads"].fillna(0).astype(float))
df["created_dt"] = pd.to_datetime(df["created"])
df["base_dt"] = pd.to_datetime(df["base_date"])
df["days_since_base"] = (df["created_dt"] - df["base_dt"]).dt.days
today = pd.Timestamp.now().normalize()
df["days_since_release"] = (today - df["created_dt"]).dt.days
df["year_half"] = df["created_dt"].apply(
lambda d: f"{d.year}H{'1' if d.month <= 6 else '2'}"
)
# Timing buckets (inclusive)
df["within_1_day"] = ((df["days_since_base"] >= 0) & (df["days_since_base"] <= 1)).astype(int)
df["within_1_week"] = ((df["days_since_base"] >= 0) & (df["days_since_base"] <= 7)).astype(int)
df["within_1_month"] = ((df["days_since_base"] >= 0) & (df["days_since_base"] <= 30)).astype(int)
# Size buckets
def size_bucket(size_b):
if pd.isna(size_b):
return "unknown"
if size_b < 5:
return "<5B"
elif size_b < 15:
return "5-15B"
else:
return "15-50B" # Includes >50B models too
df["size_bucket"] = df["model_size_b"].apply(size_bucket)
df = df.sort_values("created_dt").reset_index(drop=True)
prior_any = []
prior_same_fmt = []
total_variants_count = []
days_since_prior_variant = []
for _, row in df.iterrows():
same_base = df[
(df["norm_base"] == row["norm_base"])
& (df["created_dt"] < row["created_dt"])
]
prior_any.append(1 if len(same_base) > 0 else 0)
same_fmt = same_base[same_base["quant_format"] == row["quant_format"]]
prior_same_fmt.append(1 if len(same_fmt) > 0 else 0)
# Count total variants for this base model across both hubs (at time of release)
all_same_base = df[df["norm_base"] == row["norm_base"]]
total_variants_count.append(len(all_same_base))
# Calculate days since prior variant was published (0 if first variant)
if len(same_base) > 0:
most_recent_prior = same_base.iloc[-1] # Already sorted by created_dt
days_diff = (row["created_dt"] - most_recent_prior["created_dt"]).days
days_since_prior_variant.append(days_diff)
else:
days_since_prior_variant.append(0)
df["prior_variant_exists"] = prior_any
df["prior_same_format_exists"] = prior_same_fmt
df["total_variants_count"] = total_variants_count
df["days_since_prior_variant"] = days_since_prior_variant
return df
def build_dataset():
print("Fetching RedHatAI models...")
rh_all = fetch_models("RedHatAI")
print(f" {len(rh_all)} total")
print("Fetching Nvidia models...")
nv_all = fetch_models("nvidia")
print(f" {len(nv_all)} total")
rows = []
for hub, models in [("RedHatAI", rh_all), ("nvidia", nv_all)]:
for m in models:
name = m["id"].split("/")[-1]
if not is_quantized(name):
continue
rows.append({
"hub": hub,
"name": name,
"created": (m.get("createdAt") or "")[:10],
"downloads": m.get("downloadsAllTime", m.get("downloads", 0)),
"downloads_30d": m.get("downloads", 0),
"quant_format": extract_quant_format(name),
"model_size_b": extract_model_size_b(name),
"norm_base": normalize_base(extract_base(name)),
})
df = pd.DataFrame(rows)
df = df[df["created"] != ""]
df = df.dropna(subset=["model_size_b"])
cache, failed = _resolve_bases(df)
df["base_date"] = df["norm_base"].map(lambda b: cache.get(b, (None, None))[0])
df["base_downloads"] = df["norm_base"].map(lambda b: cache.get(b, (None, None))[1])
# Keep rows with failed resolution in CSV (base_date="FAILED") but exclude from training
df.loc[df["base_date"].isna(), "base_date"] = "FAILED"
df.to_csv(CSV_PATH, index=False)
n_failed = (df["base_date"] == "FAILED").sum()
print(f" Saved {len(df)} rows to {CSV_PATH} ({n_failed} with FAILED base_date)")
df = df[df["base_date"] != "FAILED"]
return _add_derived_columns(df)
def retry_failed():
"""Load CSV, re-resolve only FAILED base models, save back."""
if not os.path.exists(CSV_PATH):
print(f"No {CSV_PATH} found. Run without --retry first.")
sys.exit(1)
df = pd.read_csv(CSV_PATH)
failed_mask = df["base_date"] == "FAILED"
failed_bases = df.loc[failed_mask, "norm_base"].unique()
if len(failed_bases) == 0:
print("No FAILED rows to retry.")
else:
print(f"Retrying {len(failed_bases)} failed base models...")
cache = {}
for i, base in enumerate(sorted(failed_bases), 1):
sample = df[df["norm_base"] == base].iloc[0]["name"]
resolve_original_info(base, sample, cache)
date, dl = cache.get(base, (None, None))
status = date if date else "STILL FAILED"
sys.stdout.write(f"\r [{i}/{len(failed_bases)}] {base}: {status} ")
sys.stdout.flush()
print()
fixed = 0
for base, (date, dl) in cache.items():
if date:
mask = df["norm_base"] == base
df.loc[mask, "base_date"] = date
df.loc[mask, "base_downloads"] = dl
fixed += mask.sum()
still_failed = (df["base_date"] == "FAILED").sum()
print(f" Fixed {fixed} rows, {still_failed} still FAILED")
df.to_csv(CSV_PATH, index=False)
print(f" Saved to {CSV_PATH}")
df = df[df["base_date"] != "FAILED"]
return _add_derived_columns(df)
# ---------------------------------------------------------------------------
# Train and report
# ---------------------------------------------------------------------------
def train_and_report(df):
features = df[["hub", "days_since_base", "days_since_release", "quant_format", "model_size_b", "log_base_downloads"]].copy()
y = df["downloads"].values.astype(float)
# Log-transform downloads (avoids heavy right skew dominating the fit)
y_log = np.log1p(y)
# One-hot encode categoricals
cat_cols = ["hub", "quant_format"]
enc = OneHotEncoder(sparse_output=False, drop="first", handle_unknown="infrequent_if_exist", min_frequency=5)
cat_encoded = enc.fit_transform(features[cat_cols])
cat_names = enc.get_feature_names_out(cat_cols)
num_cols = ["days_since_base", "days_since_release", "model_size_b", "log_base_downloads"]
num_data = features[num_cols].values.astype(float)
X = np.hstack([num_data, cat_encoded])
col_names = list(num_cols) + list(cat_names)
model = LinearRegression()
model.fit(X, y_log)
r2 = model.score(X, y_log)
print(f"\n{'=' * 70}")
print(f" Model 1: All models — hub, format, size, base popularity")
print(f" R² = {r2:.4f} (n = {len(y)})")
print(f"{'=' * 70}")
print(f"\n{'Feature':<40} {'Coefficient':>12} Interpretation")
print("-" * 80)
print(f"{'(intercept)':<40} {model.intercept_:>12.4f} baseline log-downloads")
order = np.argsort(np.abs(model.coef_))[::-1]
for i in order:
coef = model.coef_[i]
name = col_names[i]
pct = (np.exp(coef) - 1) * 100
if name in num_cols:
interp = f"{pct:+.2f}% downloads per unit increase"
else:
interp = f"{pct:+.1f}% vs baseline category"
print(f"{name:<40} {coef:>12.4f} {interp}")
# Summary stats per format
print(f"\n{'=' * 70}")
print(f" Mean downloads by quant format")
print(f"{'=' * 70}")
fmt_stats = df.groupby("quant_format")["downloads"].agg(["count", "mean", "median"])
fmt_stats = fmt_stats.sort_values("mean", ascending=False)
fmt_stats["mean"] = fmt_stats["mean"].astype(int)
fmt_stats["median"] = fmt_stats["median"].astype(int)
print(fmt_stats.to_string())
return model, enc, col_names
def train_format_trend_model(df):
"""Model 2: RedHatAI only, with format × year_half interaction."""
df2 = df[df["hub"] == "RedHatAI"].copy()
df2["format_x_period"] = df2["quant_format"] + ":" + df2["year_half"]
features = df2[["days_since_base", "model_size_b", "log_base_downloads", "format_x_period", "within_1_day", "within_1_week", "within_1_month"]].copy()
y = df2["downloads"].values.astype(float)
y_log = np.log1p(y)
cat_cols = ["format_x_period"]
enc = OneHotEncoder(sparse_output=False, drop="first", handle_unknown="infrequent_if_exist", min_frequency=3)
cat_encoded = enc.fit_transform(features[cat_cols])
cat_names = enc.get_feature_names_out(cat_cols)
num_cols = ["days_since_base", "model_size_b", "log_base_downloads", "within_1_day", "within_1_week", "within_1_month"]
num_data = features[num_cols].values.astype(float)
X = np.hstack([num_data, cat_encoded])
col_names = list(num_cols) + list(cat_names)
model = LinearRegression()
model.fit(X, y_log)
r2 = model.score(X, y_log)
print(f"\n{'=' * 70}")
print(f" Model 2: RedHatAI only — format × period interaction")
print(f" R² = {r2:.4f} (n = {len(y)})")
print(f"{'=' * 70}")
print(f"\n{'Feature':<50} {'Coef':>8} {'Multiplier':>10} Interpretation")
print("-" * 100)
print(f"{'(intercept)':<50} {model.intercept_:>8.3f} {'':>10} baseline")
order = np.argsort(np.abs(model.coef_))[::-1]
for i in order:
coef = model.coef_[i]
name = col_names[i]
mult = np.exp(coef)
if name in num_cols:
if name == "within_1_day":
interp = f"{'×'}{mult:.2f} when released within 1 day of base"
elif name == "within_1_week":
interp = f"{'×'}{mult:.2f} when released within 1 week of base"
elif name == "within_1_month":
interp = f"{'×'}{mult:.2f} when released within 1 month of base"
else:
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per unit"
else:
interp = f"{'×'}{mult:.2f} vs baseline"
print(f"{name:<50} {coef:>8.3f} {mult:>10.2f} {interp}")
# Extract just the format×period interactions and pivot into a readable table
interactions = {}
for i, name in enumerate(col_names):
if name.startswith("format_x_period_"):
tag = name.replace("format_x_period_", "")
if ":" in tag:
fmt, period = tag.rsplit(":", 1)
interactions.setdefault(fmt, {})[period] = model.coef_[i]
if interactions:
periods = sorted({p for fmt_dict in interactions.values() for p in fmt_dict})
formats = sorted(interactions.keys())
print(f"\n{'=' * 70}")
print(f" Format interest over time (coefficient = log-multiplier on downloads)")
print(f" Positive = more downloads than format average, negative = fewer")
print(f"{'=' * 70}")
header = f"{'Format':<15}" + "".join(f"{p:>10}" for p in periods)
print(header)
print("-" * len(header))
for fmt in formats:
vals = interactions[fmt]
row = f"{fmt:<15}"
for p in periods:
if p in vals:
row += f"{vals[p]:>10.2f}"
else:
row += f"{'—':>10}"
print(row)
# Also show as multiplier (more intuitive)
print(f"\n{'=' * 70}")
print(f" Same table as download multiplier (1.00 = neutral)")
print(f"{'=' * 70}")
print(header)
print("-" * len(header))
for fmt in formats:
vals = interactions[fmt]
row = f"{fmt:<15}"
for p in periods:
if p in vals:
row += f"{np.exp(vals[p]):>10.2f}"
else:
row += f"{'—':>10}"
print(row)
# Show sample counts per format×period
print(f"\n{'=' * 70}")
print(f" Sample counts per format × period")
print(f"{'=' * 70}")
ct = pd.crosstab(df2["quant_format"], df2["year_half"])
print(ct.to_string())
return model, enc, col_names
def train_first_mover_model(df):
"""Model 3: RedHatAI only, prior variant (no interactions, no timing buckets)."""
df3 = df[df["hub"] == "RedHatAI"].copy()
features = df3[[
"days_since_base",
"model_size_b", "quant_format", "prior_variant_exists",
"prior_same_format_exists", "log_base_downloads",
"total_variants_count", "days_since_prior_variant",
]].copy()
y = df3["downloads"].values.astype(float)
y_log = np.log1p(y)
cat_cols = ["quant_format"]
enc = OneHotEncoder(sparse_output=False, drop="first", handle_unknown="infrequent_if_exist", min_frequency=3)
cat_encoded = enc.fit_transform(features[cat_cols])
cat_names = enc.get_feature_names_out(cat_cols)
num_cols = ["days_since_base", "model_size_b", "prior_variant_exists", "prior_same_format_exists", "log_base_downloads", "total_variants_count", "days_since_prior_variant"]
num_data = features[num_cols].values.astype(float)
X = np.hstack([num_data, cat_encoded])
col_names = list(num_cols) + list(cat_names)
model = LinearRegression()
model.fit(X, y_log)
r2 = model.score(X, y_log)
print(f"\n{'=' * 70}")
print(f" Model 3: RedHatAI only — prior variant exists (no interactions, no timing)")
print(f" R² = {r2:.4f} (n = {len(y)})")
print(f"{'=' * 70}")
print(f"\n{'Feature':<50} {'Coef':>8} {'Multiplier':>10} Interpretation")
print("-" * 100)
print(f"{'(intercept)':<50} {model.intercept_:>8.3f} {'':>10} baseline")
order = np.argsort(np.abs(model.coef_))[::-1]
for i in order:
coef = model.coef_[i]
name = col_names[i]
mult = np.exp(coef)
if name in num_cols:
if name == "prior_variant_exists":
interp = f"{'×'}{mult:.2f} when any prior variant exists"
elif name == "prior_same_format_exists":
interp = f"{'×'}{mult:.2f} when same format already released"
elif name == "total_variants_count":
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per additional variant (RH+nvidia)"
elif name == "days_since_prior_variant":
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per day since prior variant published"
else:
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per unit"
else:
interp = f"{'×'}{mult:.2f} vs baseline"
print(f"{name:<50} {coef:>8.3f} {mult:>10.2f} {interp}")
# Show breakdowns
print(f"\n{'=' * 70}")
print(f" Downloads by first-mover status (RedHatAI)")
print(f"{'=' * 70}")
for val in [0, 1]:
sub = df3[df3["prior_variant_exists"] == val]
label = "First quant variant for base" if val == 0 else "Prior variant existed"
print(f" {label}: n={len(sub)}, mean={sub['downloads'].mean():,.0f}, median={sub['downloads'].median():,.0f}")
print()
for val in [0, 1]:
sub = df3[df3["prior_same_format_exists"] == val]
label = "First of this quant format for base" if val == 0 else "Same format already existed"
print(f" {label}: n={len(sub)}, mean={sub['downloads'].mean():,.0f}, median={sub['downloads'].median():,.0f}")
return model, enc, col_names
def train_format_size_model(df):
"""Model 4: RedHatAI only, format × size + prior variant exists + timing buckets."""
df4 = df[df["hub"] == "RedHatAI"].copy()
df4["format_x_size"] = df4["quant_format"] + ":" + df4["size_bucket"]
features = df4[[
"days_since_base",
"format_x_size", "prior_variant_exists",
"prior_same_format_exists", "log_base_downloads",
"within_1_day", "within_1_week", "within_1_month",
"total_variants_count", "days_since_prior_variant",
]].copy()
y = df4["downloads"].values.astype(float)
y_log = np.log1p(y)
cat_cols = ["format_x_size"]
enc = OneHotEncoder(sparse_output=False, drop="first", handle_unknown="infrequent_if_exist", min_frequency=3)
cat_encoded = enc.fit_transform(features[cat_cols])
cat_names = enc.get_feature_names_out(cat_cols)
num_cols = ["days_since_base", "prior_variant_exists", "prior_same_format_exists", "log_base_downloads", "within_1_day", "within_1_week", "within_1_month", "total_variants_count", "days_since_prior_variant"]
num_data = features[num_cols].values.astype(float)
X = np.hstack([num_data, cat_encoded])
col_names = list(num_cols) + list(cat_names)
model = LinearRegression()
model.fit(X, y_log)
r2 = model.score(X, y_log)
print(f"\n{'=' * 70}")
print(f" Model 4: RedHatAI only — format × size + prior variant exists + timing")
print(f" R² = {r2:.4f} (n = {len(y)})")
print(f"{'=' * 70}")
print(f"\n{'Feature':<50} {'Coef':>8} {'Multiplier':>10} Interpretation")
print("-" * 100)
print(f"{'(intercept)':<50} {model.intercept_:>8.3f} {'':>10} baseline")
order = np.argsort(np.abs(model.coef_))[::-1]
for i in order:
coef = model.coef_[i]
name = col_names[i]
mult = np.exp(coef)
if name in num_cols:
if name == "prior_variant_exists":
interp = f"{'×'}{mult:.2f} when any prior variant exists"
elif name == "prior_same_format_exists":
interp = f"{'×'}{mult:.2f} when same format already released"
elif name == "within_1_day":
interp = f"{'×'}{mult:.2f} when released within 1 day of base"
elif name == "within_1_week":
interp = f"{'×'}{mult:.2f} when released within 1 week of base"
elif name == "within_1_month":
interp = f"{'×'}{mult:.2f} when released within 1 month of base"
elif name == "total_variants_count":
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per additional variant (RH+nvidia)"
elif name == "days_since_prior_variant":
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per day since prior variant published"
else:
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per unit"
else:
interp = f"{'×'}{mult:.2f} vs baseline"
print(f"{name:<50} {coef:>8.3f} {mult:>10.2f} {interp}")
# Extract just the format×size interactions and pivot into a readable table
interactions = {}
for i, name in enumerate(col_names):
if name.startswith("format_x_size_"):
tag = name.replace("format_x_size_", "")
if ":" in tag:
fmt, size = tag.rsplit(":", 1)
interactions.setdefault(fmt, {})[size] = model.coef_[i]
if interactions:
sizes = ["<5B", "5-15B", "15-50B"] # Fixed order
formats = sorted(interactions.keys())
print(f"\n{'=' * 70}")
print(f" Format effectiveness by model size (coefficient = log-multiplier on downloads)")
print(f" Positive = more downloads than format average, negative = fewer")
print(f"{'=' * 70}")
header = f"{'Format':<15}" + "".join(f"{s:>10}" for s in sizes)
print(header)
print("-" * len(header))
for fmt in formats:
vals = interactions[fmt]
row = f"{fmt:<15}"
for s in sizes:
if s in vals:
row += f"{vals[s]:>10.2f}"
else:
row += f"{'—':>10}"
print(row)
# Also show as multiplier (more intuitive)
print(f"\n{'=' * 70}")
print(f" Same table as download multiplier (1.00 = neutral)")
print(f"{'=' * 70}")
print(header)
print("-" * len(header))
for fmt in formats:
vals = interactions[fmt]
row = f"{fmt:<15}"
for s in sizes:
if s in vals:
row += f"{np.exp(vals[s]):>10.2f}"
else:
row += f"{'—':>10}"
print(row)
# Show sample counts per format×size
print(f"\n{'=' * 70}")
print(f" Sample counts per format × size")
print(f"{'=' * 70}")
ct = pd.crosstab(df4["quant_format"], df4["size_bucket"])
print(ct.to_string())
# Show breakdowns
print(f"\n{'=' * 70}")
print(f" Downloads by first-mover status (RedHatAI)")
print(f"{'=' * 70}")
for val in [0, 1]:
sub = df4[df4["prior_variant_exists"] == val]
label = "First quant variant for base" if val == 0 else "Prior variant existed"
print(f" {label}: n={len(sub)}, mean={sub['downloads'].mean():,.0f}, median={sub['downloads'].median():,.0f}")
print()
for val in [0, 1]:
sub = df4[df4["prior_same_format_exists"] == val]
label = "First of this quant format for base" if val == 0 else "Same format already existed"
print(f" {label}: n={len(sub)}, mean={sub['downloads'].mean():,.0f}, median={sub['downloads'].median():,.0f}")
print(f"\n{'=' * 70}")
print(f" Downloads by release timing relative to base model (RedHatAI)")
print(f"{'=' * 70}")
timing_data = [
("Within 1 day (0-1d)", df4["within_1_day"] == 1),
("Within 1 week (2-7d)", df4["within_1_week"] == 1),
("Within 1 month (8-30d)", df4["within_1_month"] == 1),
("After 1 month (31+d)", (df4["within_1_day"] == 0) & (df4["within_1_week"] == 0) & (df4["within_1_month"] == 0)),
]
for label, mask in timing_data:
sub = df4[mask]
if len(sub) > 0:
print(f" {label}: n={len(sub)}, mean={sub['downloads'].mean():,.0f}, median={sub['downloads'].median():,.0f}")
return model, enc, col_names
def train_overlap_model(df):
"""Model 5: RedHatAI only, limited to (base, format) pairs where both RH and nvidia released within 1 week."""
# Find overlapping (norm_base, quant_format) pairs where both released within 1 week of each other
overlap_pairs = set()
for (norm_base, quant_format) in df[["norm_base", "quant_format"]].drop_duplicates().values:
rh_models = df[(df["hub"] == "RedHatAI") & (df["norm_base"] == norm_base) & (df["quant_format"] == quant_format)]
nv_models = df[(df["hub"] == "nvidia") & (df["norm_base"] == norm_base) & (df["quant_format"] == quant_format)]
if len(rh_models) > 0 and len(nv_models) > 0:
# Get release dates
rh_dates = pd.to_datetime(rh_models["created"])
nv_dates = pd.to_datetime(nv_models["created"])
# Check if any pair released within 14 days of each other
for rh_date in rh_dates:
for nv_date in nv_dates:
if abs((rh_date - nv_date).days) <= 14:
overlap_pairs.add((norm_base, quant_format))
break
if (norm_base, quant_format) in overlap_pairs:
break
# Filter to RedHatAI models where the (base, format) pair exists in both within 1 week
df5 = df[df["hub"] == "RedHatAI"].copy()
df5["pair"] = df5[["norm_base", "quant_format"]].apply(tuple, axis=1)
df5 = df5[df5["pair"].isin(overlap_pairs)].copy()
print(f"\n Model 5 uses {len(df5)} RedHatAI models (out of {len(df[df['hub'] == 'RedHatAI'])} total)")
print(f" These are models where both RedHatAI and nvidia released the same (base, format) within 2 weeks")
features = df5[[
"days_since_base",
"model_size_b", "quant_format", "prior_same_format_exists", "log_base_downloads",
]].copy()
y = df5["downloads"].values.astype(float)
y_log = np.log1p(y)
cat_cols = ["quant_format"]
enc = OneHotEncoder(sparse_output=False, drop="first", handle_unknown="infrequent_if_exist", min_frequency=3)
cat_encoded = enc.fit_transform(features[cat_cols])
cat_names = enc.get_feature_names_out(cat_cols)
num_cols = ["days_since_base", "model_size_b", "prior_same_format_exists", "log_base_downloads"]
num_data = features[num_cols].values.astype(float)
X = np.hstack([num_data, cat_encoded])
col_names = list(num_cols) + list(cat_names)
model = LinearRegression()
model.fit(X, y_log)
r2 = model.score(X, y_log)
print(f"\n{'=' * 70}")
print(f" Model 5: RedHatAI only — both RH+nvidia released within 2 weeks")
print(f" R² = {r2:.4f} (n = {len(y)})")
print(f"{'=' * 70}")
print(f"\n{'Feature':<50} {'Coef':>8} {'Multiplier':>10} Interpretation")
print("-" * 100)
print(f"{'(intercept)':<50} {model.intercept_:>8.3f} {'':>10} baseline")
order = np.argsort(np.abs(model.coef_))[::-1]
for i in order:
coef = model.coef_[i]
name = col_names[i]
mult = np.exp(coef)
if name in num_cols:
if name == "prior_same_format_exists":
interp = f"{'×'}{mult:.2f} when same format already released"
else:
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per unit"
else:
interp = f"{'×'}{mult:.2f} vs baseline"
print(f"{name:<50} {coef:>8.3f} {mult:>10.2f} {interp}")
# Show breakdowns
print(f"\n{'=' * 70}")
print(f" Downloads by first-of-format status (RedHatAI, overlap only)")
print(f"{'=' * 70}")
for val in [0, 1]:
sub = df5[df5["prior_same_format_exists"] == val]
label = "First of this quant format for base" if val == 0 else "Same format already existed"
print(f" {label}: n={len(sub)}, mean={sub['downloads'].mean():,.0f}, median={sub['downloads'].median():,.0f}")
return model, enc, col_names
def train_granular_timing_model(df):
"""Model 6: RedHatAI only, granular timing of when nvidia released same format."""
# Find overlapping (norm_base, quant_format) pairs (no timing restriction)
rh_pairs = set(df[df["hub"] == "RedHatAI"][["norm_base", "quant_format"]].apply(tuple, axis=1))
nv_pairs = set(df[df["hub"] == "nvidia"][["norm_base", "quant_format"]].apply(tuple, axis=1))
overlap_pairs = rh_pairs & nv_pairs
# Filter to RedHatAI models where the (base, format) pair exists in both
df6 = df[df["hub"] == "RedHatAI"].copy()
df6["pair"] = df6[["norm_base", "quant_format"]].apply(tuple, axis=1)
df6 = df6[df6["pair"].isin(overlap_pairs)].copy()
# For each RedHat model, determine timing relative to base and find nvidia timing
def get_timing_bucket(days):
"""Convert days_since_base to timing bucket."""
if days <= 1:
return "1day"
elif days <= 3:
return "3days"
elif days <= 7:
return "1week"
elif days <= 14:
return "2weeks"
elif days <= 30:
return "1month"
else:
return "after_1month"
rh_nv_interaction = []
for _, rh_row in df6.iterrows():
rh_timing = get_timing_bucket(rh_row["days_since_base"])
nv_models = df[(df["hub"] == "nvidia") &
(df["norm_base"] == rh_row["norm_base"]) &
(df["quant_format"] == rh_row["quant_format"])]
if len(nv_models) > 0:
# Get nvidia's days_since_base (use earliest if multiple)
nv_days_since_base = nv_models["days_since_base"].min()
nv_timing = get_timing_bucket(nv_days_since_base)
interaction = f"rh_{rh_timing}_x_nv_{nv_timing}"
else:
interaction = "no_nv_release" # Shouldn't happen due to filter
rh_nv_interaction.append(interaction)
df6["rh_nv_timing_interaction"] = rh_nv_interaction
print(f"\n Model 6 uses {len(df6)} RedHatAI models (out of {len(df[df['hub'] == 'RedHatAI'])} total)")
print(f" Limited to models where both RedHat and nvidia released the same (base, format)")
print(f"\n RH-NV timing interaction breakdown:")
print(df6["rh_nv_timing_interaction"].value_counts().sort_index())
features = df6[[
"model_size_b", "quant_format", "log_base_downloads", "rh_nv_timing_interaction",
]].copy()
y = df6["downloads"].values.astype(float)
y_log = np.log1p(y)
cat_cols = ["quant_format", "rh_nv_timing_interaction"]
enc = OneHotEncoder(sparse_output=False, drop="first", handle_unknown="infrequent_if_exist", min_frequency=1)
cat_encoded = enc.fit_transform(features[cat_cols])
cat_names = enc.get_feature_names_out(cat_cols)
num_cols = ["model_size_b", "log_base_downloads"]
num_data = features[num_cols].values.astype(float)
X = np.hstack([num_data, cat_encoded])
col_names = list(num_cols) + list(cat_names)
model = LinearRegression()
model.fit(X, y_log)
r2 = model.score(X, y_log)
print(f"\n{'=' * 70}")
print(f" Model 6: RedHatAI only — granular timing relative to nvidia release")
print(f" R² = {r2:.4f} (n = {len(y)})")
print(f"{'=' * 70}")
print(f"\n{'Feature':<50} {'Coef':>8} {'Multiplier':>10} Interpretation")
print("-" * 100)
print(f"{'(intercept)':<50} {model.intercept_:>8.3f} {'':>10} baseline")
order = np.argsort(np.abs(model.coef_))[::-1]
for i in order:
coef = model.coef_[i]
name = col_names[i]
mult = np.exp(coef)
if name in num_cols:
pct = (mult - 1) * 100
interp = f"{pct:+.2f}% per unit"
else:
interp = f"{'×'}{mult:.2f} vs baseline"
print(f"{name:<50} {coef:>8.3f} {mult:>10.2f} {interp}")
# Extract RH-NV timing interactions and pivot into a readable table
interactions = {}
for i, name in enumerate(col_names):
if name.startswith("rh_nv_timing_interaction_"):
tag = name.replace("rh_nv_timing_interaction_", "")
if "_x_nv_" in tag:
parts = tag.split("_x_nv_")
rh_timing = parts[0].replace("rh_", "")
nv_timing = parts[1]
interactions.setdefault(rh_timing, {})[nv_timing] = model.coef_[i]
if interactions:
timing_order = ["1day", "3days", "1week", "2weeks", "1month", "after_1month"]
rh_timings = [t for t in timing_order if t in interactions]
nv_timings = timing_order
print(f"\n{'=' * 90}")
print(f" RH vs NV timing interactions (coefficient = log-multiplier on downloads)")
print(f"{'=' * 90}")
header = f"{'RH timing':<15}" + "".join(f"{'nv_'+nv:>12}" for nv in nv_timings)
print(header)
print("-" * len(header))
for rh_t in rh_timings:
vals = interactions[rh_t]
row = f"{rh_t:<15}"
for nv_t in nv_timings:
if nv_t in vals:
row += f"{vals[nv_t]:>12.2f}"
else:
row += f"{'—':>12}"
print(row)
# Also show as multiplier
print(f"\n{'=' * 90}")
print(f" Same table as download multiplier (1.00 = neutral)")
print(f"{'=' * 90}")
print(header)
print("-" * len(header))
for rh_t in rh_timings:
vals = interactions[rh_t]
row = f"{rh_t:<15}"
for nv_t in nv_timings:
if nv_t in vals:
row += f"{np.exp(vals[nv_t]):>12.2f}"
else:
row += f"{'—':>12}"
print(row)
return model, enc, col_names
def main():
parser = argparse.ArgumentParser(description="Train download prediction models")
parser.add_argument("--retry", action="store_true",
help="Retry fetching only FAILED rows from CSV")
parser.add_argument("--fetch-all", action="store_true",
help="Re-fetch everything from HF API")
args = parser.parse_args()
if args.fetch_all:
df = build_dataset()
elif args.retry:
df = retry_failed()
elif os.path.exists(CSV_PATH):
# Default: use CSV without fetching
print(f"Loading data from {CSV_PATH} (use --retry for failed rows, --fetch-all to re-fetch everything)")
df = pd.read_csv(CSV_PATH)
df = df[df["base_date"] != "FAILED"]
df = _add_derived_columns(df)
else:
# No CSV exists, fetch everything
df = build_dataset()
print(f"\nDataset: {len(df)} quantized models with known base date and model size")
print(f" Hubs: {df['hub'].value_counts().to_dict()}")
print(f" Formats: {df['quant_format'].value_counts().to_dict()}")
print(f" Prior variant exists: {df['prior_variant_exists'].value_counts().to_dict()}")
train_and_report(df)
train_format_trend_model(df)
train_first_mover_model(df)
train_format_size_model(df)
train_overlap_model(df)
train_granular_timing_model(df)
if __name__ == "__main__":
main()
Loading data from hf_quant_dataset.csv...
RedHatAI quantized: 379
Nvidia quantized: 73
Download threshold: >100.0K combined
============================================================
EXACT MATCHES (16 models, >100.0K combined) [delta = RedHatAI minus Nvidia]
============================================================
Model Name | Base Date | RH Date | RH DLs | NV Date | NV DLs | Date Delta | DL Delta
----------------------------------------|------------|------------|--------|------------|--------|------------|---------
diffusiongemma-26B-A4B-it-NVFP4 | 2026-06-09 | 2026-06-09 | 606.1K | 2026-06-10 | 1.7M | -1d | -1.1M
NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 | 2026-05-26 | 2026-06-09 | 48 | 2026-06-03 | 527.3K | +6d | -527.3K
Qwen3.6-35B-A3B-NVFP4 | 2026-04-15 | 2026-04-17 | 6.5M | 2026-05-27 | 8.5M | -40d | -2.0M
Qwen3.5-122B-A10B-NVFP4 | 2026-02-24 | 2026-03-17 | 569.6K | 2026-05-13 | 93.9K | -57d | +475.7K
NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 2026-03-06 | 2026-03-26 | 5.1K | 2026-03-10 | 2.2M | +16d | -2.2M
NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 | 2026-03-06 | 2026-03-27 | 11.3K | 2026-03-10 | 5.5M | +17d | -5.5M
Qwen3-Next-80B-A3B-Thinking-NVFP4 | 2025-09-09 | 2025-12-24 | 1.7K | 2025-12-11 | 210.3K | +13d | -208.6K
Qwen3-Next-80B-A3B-Instruct-NVFP4 | 2025-09-09 | 2025-12-12 | 79.1K | 2025-12-09 | 218.1K | +3d | -139.0K
NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 | 2025-12-03 | 2025-12-23 | 9.2K | 2025-12-06 | 5.9M | +17d | -5.9M
Qwen3-14B-NVFP4 | 2025-04-27 | 2025-10-23 | 75.2K | 2025-09-09 | 549.5K | +44d | -474.3K
Qwen3-8B-NVFP4 | 2025-04-27 | 2025-10-23 | 27.8K | 2025-09-09 | 444.6K | +44d | -416.8K
Llama-3.1-8B-Instruct-NVFP4 | 2024-07-18 | 2025-10-23 | 128.1K | 2025-09-05 | 1.0M | +48d | -903.7K
Qwen3-32B-NVFP4 | 2025-04-27 | 2025-06-27 | 139.9K | 2025-09-09 | 627.2K | -74d | -487.3K
Qwen3-30B-A3B-NVFP4 | 2025-04-27 | 2025-06-17 | 190.1K | 2025-07-08 | 705.0K | -21d | -514.9K
Llama-4-Scout-17B-16E-Instruct-NVFP4 | 2025-04-02 | 2025-07-08 | 20.5K | 2025-04-14 | 584.4K | +85d | -563.9K
Llama-3.3-70B-Instruct-NVFP4 | 2024-11-26 | 2025-08-05 | 26.5K | 2025-01-16 | 588.7K | +201d | -562.2K
============================================================
FUZZY MATCHES — same base model, different quant (28 base models, >100.0K combined) [delta = RH minus NV]
============================================================
Base Model | Full Name | Date | Downloads | Date Delta | DL Delta | Variants |
-----------------------------------|--------------------------------------------------------------------------------|------------|-----------|------------|----------|-----------|----
diffusiongemma-26b-a4b | [base: diffusiongemma-26b-a4b] | 2026-06-09 | | | | |
| RedHatAI/diffusiongemma-26B-A4B-it-FP8-dynamic | 2026-06-09 | 284.5K | | | |
| RedHatAI/diffusiongemma-26B-A4B-it-NVFP4 | 2026-06-09 | 606.1K | | | |
| nvidia/diffusiongemma-26B-A4B-it-NVFP4 | 2026-06-10 | 1.7M | | | |
| ^ DELTA (earliest date, total DLs) | | | -1d | -796.9K | RH:2 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
nvidia-nemotron-3-ultra-550b-a55b | [base: nvidia-nemotron-3-ultra-550b-a55b] | 2026-05-26 | | | | |
| RedHatAI/NVIDIA-Nemotron-3-Ultra-550B-A55B-FP8-block | 2026-06-04 | 7.9K | | | |
| RedHatAI/NVIDIA-Nemotron-3-Ultra-550B-A55B-FP8-dynamic | 2026-06-04 | 14.8K | | | |
| RedHatAI/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 | 2026-06-09 | 48 | | | |
| RedHatAI/NVIDIA-Nemotron-3-Ultra-550B-A55B-quantized.w4a16 | 2026-06-04 | 2.4K | | | |
| nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 | 2026-06-03 | 527.3K | | | |
| ^ DELTA (earliest date, total DLs) | | | +1d | -502.2K | RH:4 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3.6-27b | [base: qwen3.6-27b] | 2026-04-21 | | | | |
| RedHatAI/Qwen3.6-27B-FP8 | 2026-05-01 | 55.0K | | | |
| nvidia/Qwen3.6-27B-NVFP4 | 2026-06-22 | 538.7K | | | |
| ^ DELTA (earliest date, total DLs) | | | -52d | -483.7K | RH:1 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3.6-35b-a3b | [base: qwen3.6-35b-a3b] | 2026-04-15 | | | | |
| RedHatAI/Qwen3.6-35B-A3B-FP8 | 2026-05-01 | 26.1K | | | |
| RedHatAI/Qwen3.6-35B-A3B-FP8-dynamic | 2026-04-27 | 27.1K | | | |
| RedHatAI/Qwen3.6-35B-A3B-NVFP4 | 2026-04-17 | 6.5M | | | |
| nvidia/Qwen3.6-35B-A3B-NVFP4 | 2026-05-27 | 8.5M | | | |
| ^ DELTA (earliest date, total DLs) | | | -40d | -2.0M | RH:3 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
gemma-4-26b-a4b | [base: gemma-4-26b-a4b] | 2026-03-11 | | | | |
| RedHatAI/gemma-4-26B-A4B-it-FP8-Dynamic | 2026-04-06 | 2.7M | | | |
| RedHatAI/gemma-4-26B-A4B-it-NVFP4 | 2026-04-05 | 2.6M | | | |
| nvidia/Gemma-4-26B-A4B-NVFP4 | 2026-05-01 | 3.8M | | | |
| ^ DELTA (earliest date, total DLs) | | | -26d | +1.5M | RH:2 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
gemma-4-31b | [base: gemma-4-31b] | 2026-03-11 | | | | |
| RedHatAI/gemma-4-31B-it-FP8-block | 2026-04-03 | 4.4M | | | |
| RedHatAI/gemma-4-31B-it-FP8-dynamic | 2026-04-06 | 792.7K | | | |
| RedHatAI/gemma-4-31B-it-NVFP4 | 2026-04-03 | 742.7K | | | |
| nvidia/Gemma-4-31B-IT-NVFP4 | 2026-04-02 | 6.6M | | | |
| ^ DELTA (earliest date, total DLs) | | | +1d | -740.5K | RH:3 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
nvidia-nemotron-3-super-120b-a12b | [base: nvidia-nemotron-3-super-120b-a12b] | 2026-03-06 | | | | |
| RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 2026-03-26 | 5.1K | | | |
| RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-FP8-Dflash | 2026-06-01 | 138 | | | |
| RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 | 2026-03-27 | 11.3K | | | |
| nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8 | 2026-03-10 | 2.2M | | | |
| nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 | 2026-03-10 | 5.5M | | | |
| ^ DELTA (earliest date, total DLs) | | | +16d | -7.8M | RH:3 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3.5-122b-a10b | [base: qwen3.5-122b-a10b] | 2026-02-24 | | | | |
| RedHatAI/Qwen3.5-122B-A10B-FP8-dynamic | 2026-03-03 | 7.6K | | | |
| RedHatAI/Qwen3.5-122B-A10B-NVFP4 | 2026-03-17 | 569.6K | | | |
| nvidia/Qwen3.5-122B-A10B-NVFP4 | 2026-05-13 | 93.9K | | | |
| ^ DELTA (earliest date, total DLs) | | | -71d | +483.3K | RH:2 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3.5-397b-a17b | [base: qwen3.5-397b-a17b] | 2026-02-16 | | | | |
| RedHatAI/Qwen3.5-397B-A17B-FP8-dynamic | 2026-02-18 | 11.8K | | | |
| nvidia/Qwen3.5-397B-A17B-NVFP4 | 2026-02-16 | 2.4M | | | |
| nvidia/Qwen3.5-397B-A17B-NVFP4-V2 | 2026-06-29 | 19.3K | | | |
| ^ DELTA (earliest date, total DLs) | | | +2d | -2.4M | RH:1 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-next-80b-a3b-thinking | [base: qwen3-next-80b-a3b-thinking] | 2025-09-09 | | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-block | 2025-12-24 | 138 | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-dynamic | 2025-12-24 | 1.1K | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Thinking-NVFP4 | 2025-12-24 | 1.7K | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Thinking-quantized.w4a16 | 2025-12-24 | 572 | | | |
| nvidia/Qwen3-Next-80B-A3B-Thinking-NVFP4 | 2025-12-11 | 210.3K | | | |
| ^ DELTA (earliest date, total DLs) | | | +13d | -206.8K | RH:4 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-next-80b-a3b-instruct | [base: qwen3-next-80b-a3b-instruct] | 2025-09-09 | | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8 | 2025-12-24 | 5.7K | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-block | 2025-12-12 | 87 | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-dynamic | 2025-12-12 | 214 | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Instruct-NVFP4 | 2025-12-12 | 79.1K | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w4a16 | 2025-12-12 | 2.9K | | | |
| RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8 | 2026-03-03 | 455 | | | |
| nvidia/Qwen3-Next-80B-A3B-Instruct-NVFP4 | 2025-12-09 | 218.1K | | | |
| ^ DELTA (earliest date, total DLs) | | | +3d | -129.7K | RH:6 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
nvidia-nemotron-3-nano-30b-a3b | [base: nvidia-nemotron-3-nano-30b-a3b] | 2025-12-03 | | | | |
| RedHatAI/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 | 2025-12-23 | 9.2K | | | |
| nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 | 2025-12-06 | 5.9M | | | |
| nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4 | 2025-12-20 | 3.7M | | | |
| ^ DELTA (earliest date, total DLs) | | | +17d | -9.6M | RH:1 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-vl-235b-a22b-instruct | [base: qwen3-vl-235b-a22b-instruct] | 2025-09-22 | | | | |
| RedHatAI/Qwen3-VL-235B-A22B-Instruct-FP8-block | 2025-09-28 | 1.2K | | | |
| RedHatAI/Qwen3-VL-235B-A22B-Instruct-FP8-dynamic | 2025-09-28 | 29.6K | | | |
| RedHatAI/Qwen3-VL-235B-A22B-Instruct-NVFP4 | 2025-10-14 | 55.9K | | | |
| nvidia/Qwen3-VL-235B-A22B-Instruct-NVFP4 | 2025-12-25 | 16.2K | | | |
| nvidia/Qwen3-VL-235B-A22B-Instruct-NVFP4-MLPerf-Inference-Closed-V6.0 | 2026-01-27 | 190.0K | | | |
| nvidia/Qwen3-VL-235B-A22B-Instruct-NVFP4-MLPerf-Inference-Closed-V6.1 | 2026-04-07 | 6.1K | | | |
| nvidia/Qwen3-VL-235B-A22B-Instruct-NVFP4-MLPerf-Inference-Closed-V6.1-FP8-KV | 2026-06-15 | 18.9K | | | |
| ^ DELTA (earliest date, total DLs) | | | -88d | -144.4K | RH:3 NV:4 |
--- | --- | --- | --- | --- | --- | --- | ---
nvidia-nemotron-nano-9b-v2 | [base: nvidia-nemotron-nano-9b-v2] | 2025-08-12 | | | | |
| RedHatAI/NVIDIA-Nemotron-Nano-9B-v2-FP8-dynamic | 2025-09-30 | 33.5K | | | |
| RedHatAI/NVIDIA-Nemotron-Nano-9B-v2-quantized.w4a16 | 2025-10-22 | 8.3K | | | |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2-FP8 | 2025-09-22 | 873.5K | | | |
| nvidia/NVIDIA-Nemotron-Nano-9B-v2-NVFP4 | 2025-10-07 | 100.7K | | | |
| ^ DELTA (earliest date, total DLs) | | | +8d | -932.3K | RH:2 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-235b-a22b | [base: qwen3-235b-a22b] | 2025-04-27 | | | | |
| RedHatAI/Qwen3-235B-A22B-FP8-dynamic | 2025-05-04 | 31.0K | | | |
| RedHatAI/Qwen3-235B-A22B-NVFP4 | 2025-10-29 | 1.3K | | | |
| nvidia/Qwen3-235B-A22B-FP8 | 2025-07-08 | 11.5K | | | |
| nvidia/Qwen3-235B-A22B-NVFP4 | 2025-07-08 | 86.9K | | | |
| ^ DELTA (earliest date, total DLs) | | | -65d | -66.0K | RH:2 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-30b-a3b | [base: qwen3-30b-a3b] | 2025-04-27 | | | | |
| RedHatAI/Qwen3-30B-A3B-FP8-block | 2025-07-01 | 96.2K | | | |
| RedHatAI/Qwen3-30B-A3B-FP8-dynamic | 2025-05-03 | 200.2K | | | |
| RedHatAI/Qwen3-30B-A3B-NVFP4 | 2025-06-17 | 190.1K | | | |
| RedHatAI/Qwen3-30B-A3B-quantized.w4a16 | 2025-05-06 | 81.6K | | | |
| nvidia/Qwen3-30B-A3B-NVFP4 | 2025-07-08 | 705.0K | | | |
| ^ DELTA (earliest date, total DLs) | | | -66d | -136.9K | RH:4 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-8b | [base: qwen3-8b] | 2025-04-27 | | | | |
| RedHatAI/Qwen3-8B-FP8-block | 2025-10-12 | 936 | | | |
| RedHatAI/Qwen3-8B-FP8-dynamic | 2025-05-02 | 277.8K | | | |
| RedHatAI/Qwen3-8B-NVFP4 | 2025-10-23 | 27.8K | | | |
| RedHatAI/Qwen3-8B-quantized.w4a16 | 2025-05-06 | 161.0K | | | |
| nvidia/Qwen3-8B-FP8 | 2025-09-09 | 93.1K | | | |
| nvidia/Qwen3-8B-NVFP4 | 2025-09-09 | 444.6K | | | |
| ^ DELTA (earliest date, total DLs) | | | -130d | -70.1K | RH:4 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-32b | [base: qwen3-32b] | 2025-04-27 | | | | |
| RedHatAI/Qwen3-32B-FP8-block | 2025-10-12 | 173 | | | |
| RedHatAI/Qwen3-32B-FP8-dynamic | 2025-05-02 | 136.8K | | | |
| RedHatAI/Qwen3-32B-NVFP4 | 2025-06-27 | 139.9K | | | |
| RedHatAI/Qwen3-32B-NVFP4A16 | 2025-06-27 | 36.5K | | | |
| RedHatAI/Qwen3-32B-quantized.w4a16 | 2025-05-05 | 51.4K | | | |
| nvidia/Qwen3-32B-NVFP4 | 2025-09-09 | 627.2K | | | |
| ^ DELTA (earliest date, total DLs) | | | -130d | -262.5K | RH:5 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen3-14b | [base: qwen3-14b] | 2025-04-27 | | | | |
| RedHatAI/Qwen3-14B-FP8-block | 2025-10-12 | 262 | | | |
| RedHatAI/Qwen3-14B-FP8-dynamic | 2025-05-02 | 11.8K | | | |
| RedHatAI/Qwen3-14B-NVFP4 | 2025-10-23 | 75.2K | | | |
| RedHatAI/Qwen3-14B-quantized.w4a16 | 2025-05-05 | 11.1K | | | |
| nvidia/Qwen3-14B-FP8 | 2025-09-09 | 48.1K | | | |
| nvidia/Qwen3-14B-NVFP4 | 2025-09-09 | 549.5K | | | |
| ^ DELTA (earliest date, total DLs) | | | -130d | -499.3K | RH:4 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
llama-4-maverick-17b-128e-instruct | [base: llama-4-maverick-17b-128e-instruct] | 2025-04-01 | | | | |
| RedHatAI/Llama-4-Maverick-17B-128E-Instruct-FP8 | 2025-05-12 | 58.8K | | | |
| RedHatAI/Llama-4-Maverick-17B-128E-Instruct-FP8-block | 2025-10-27 | 139 | | | |
| RedHatAI/Llama-4-Maverick-17B-128E-Instruct-NVFP4 | 2025-10-27 | 10.0K | | | |
| RedHatAI/Llama-4-Maverick-17B-128E-Instruct-quantized.w4a16 | 2025-06-12 | 23.0K | | | |
| nvidia/Llama-4-Maverick-17B-128E-Instruct-FP8 | 2025-04-14 | 15.4K | | | |
| ^ DELTA (earliest date, total DLs) | | | +28d | +76.6K | RH:4 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
llama-4-scout-17b-16e-instruct | [base: llama-4-scout-17b-16e-instruct] | 2025-04-02 | | | | |
| RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-block | 2025-10-23 | 724 | | | |
| RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic | 2025-04-10 | 213.1K | | | |
| RedHatAI/Llama-4-Scout-17B-16E-Instruct-NVFP4 | 2025-07-08 | 20.5K | | | |
| RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16 | 2025-04-25 | 672.4K | | | |
| nvidia/Llama-4-Scout-17B-16E-Instruct-FP8 | 2025-04-14 | 2.0M | | | |
| nvidia/Llama-4-Scout-17B-16E-Instruct-NVFP4 | 2025-04-14 | 584.4K | | | |
| ^ DELTA (earliest date, total DLs) | | | -4d | -1.7M | RH:4 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
qwen2.5-vl-7b-instruct | [base: qwen2.5-vl-7b-instruct] | 2025-01-26 | | | | |
| RedHatAI/Qwen2.5-VL-7B-Instruct-FP8-Dynamic | 2025-02-06 | 338.1K | | | |
| RedHatAI/Qwen2.5-VL-7B-Instruct-quantized.w4a16 | 2025-02-07 | 44.5K | | | |
| RedHatAI/Qwen2.5-VL-7B-Instruct-quantized.w8a8 | 2025-02-07 | 100.2K | | | |
| nvidia/Qwen2.5-VL-7B-Instruct-FP8 | 2025-09-10 | 11.7K | | | |
| nvidia/Qwen2.5-VL-7B-Instruct-NVFP4 | 2025-09-10 | 392.2K | | | |
| ^ DELTA (earliest date, total DLs) | | | -216d | +79.0K | RH:3 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
llama-3.3-70b-instruct | [base: llama-3.3-70b-instruct] | 2024-11-26 | | | | |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-block | 2025-10-20 | 25.8K | | | |
| RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic | 2024-12-11 | 2.9M | | | |
| RedHatAI/Llama-3.3-70B-Instruct-NVFP4 | 2025-08-05 | 26.5K | | | |
| RedHatAI/Llama-3.3-70B-Instruct-quantized.w4a16 | 2025-01-02 | 97.9K | | | |
| RedHatAI/Llama-3.3-70B-Instruct-quantized.w8a8 | 2025-01-20 | 207.2K | | | |
| nvidia/Llama-3.3-70B-Instruct-FP8 | 2025-05-05 | 652.4K | | | |
| nvidia/Llama-3.3-70B-Instruct-NVFP4 | 2025-01-16 | 588.7K | | | |
| ^ DELTA (earliest date, total DLs) | | | -36d | +2.0M | RH:5 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
llama-3.2-3b-instruct | [base: llama-3.2-3b-instruct] | 2024-09-18 | | | | |
| RedHatAI/Llama-3.2-3B-Instruct-FP8 | 2024-09-26 | 339.1K | | | |
| RedHatAI/Llama-3.2-3B-Instruct-FP8-dynamic | 2024-09-25 | 78.9K | | | |
| RedHatAI/Llama-3.2-3B-Instruct-quantized.w8a8 | 2024-09-25 | 64.6K | | | |
| nvidia/Meta-Llama-3.2-3B-Instruct-ONNX-INT4 | 2024-11-15 | 0 | | | |
| ^ DELTA (earliest date, total DLs) | | | -51d | +482.5K | RH:3 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
llama-3.1-405b-instruct | [base: llama-3.1-405b-instruct] | 2024-07-16 | | | | |
| RedHatAI/Meta-Llama-3.1-405B-Instruct-FP8 | 2024-07-23 | 44.5K | | | |
| RedHatAI/Meta-Llama-3.1-405B-Instruct-FP8-dynamic | 2024-07-23 | 70.3K | | | |
| RedHatAI/Meta-Llama-3.1-405B-Instruct-quantized.w4a16 | 2024-08-09 | 23.8K | | | |
| RedHatAI/Meta-Llama-3.1-405B-Instruct-quantized.w8a16 | 2024-08-19 | 2.0K | | | |
| RedHatAI/Meta-Llama-3.1-405B-Instruct-quantized.w8a8 | 2024-08-19 | 4.2K | | | |
| nvidia/Llama-3.1-405B-Instruct-FP8 | 2024-08-29 | 106.1K | | | |
| nvidia/Llama-3.1-405B-Instruct-NVFP4 | 2025-01-16 | 63.2K | | | |
| ^ DELTA (earliest date, total DLs) | | | -37d | -24.5K | RH:5 NV:2 |
--- | --- | --- | --- | --- | --- | --- | ---
llama-3.1-70b-instruct | [base: llama-3.1-70b-instruct] | 2024-07-16 | | | | |
| RedHatAI/Llama-3.1-70B-Instruct-NVFP4 | 2025-06-27 | 5.0K | | | |
| RedHatAI/Llama-3.1-70B-Instruct-NVFP4A16 | 2025-06-27 | 183 | | | |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8 | 2024-07-23 | 1.4M | | | |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-FP8-dynamic | 2024-07-23 | 96.0K | | | |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 | 2024-07-31 | 1.8M | | | |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w8a16 | 2024-07-24 | 14.5K | | | |
| RedHatAI/Meta-Llama-3.1-70B-Instruct-quantized.w8a8 | 2024-07-29 | 225.4K | | | |
| nvidia/Llama-3.1-70B-Instruct-FP8 | 2024-08-29 | 119.6K | | | |
| ^ DELTA (earliest date, total DLs) | | | -37d | +3.5M | RH:7 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
llama-3.1-8b-instruct | [base: llama-3.1-8b-instruct] | 2024-07-18 | | | | |
| RedHatAI/Llama-3.1-8B-Instruct-FP8-block | 2025-10-10 | 724 | | | |
| RedHatAI/Llama-3.1-8B-Instruct-NVFP4 | 2025-10-23 | 128.1K | | | |
| RedHatAI/Meta-Llama-3.1-8B-Instruct-FP8 | 2024-07-23 | 6.3M | | | |
| RedHatAI/Meta-Llama-3.1-8B-Instruct-FP8-dynamic | 2024-07-23 | 729.0K | | | |
| RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w4a16 | 2024-07-26 | 1.5M | | | |
| RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a16 | 2024-07-24 | 94.4K | | | |
| RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8 | 2024-07-24 | 424.0K | | | |
| nvidia/Llama-3.1-8B-Instruct-FP8 | 2024-08-29 | 2.3M | | | |
| nvidia/Llama-3.1-8B-Instruct-NVFP4 | 2025-09-05 | 1.0M | | | |
| nvidia/Meta-Llama-3.1-8B-Instruct-ONNX-INT4 | 2024-11-15 | 2.0K | | | |
| ^ DELTA (earliest date, total DLs) | | | -37d | +5.8M | RH:7 NV:3 |
--- | --- | --- | --- | --- | --- | --- | ---
mistral-7b-instruct-v0.3 | [base: mistral-7b-instruct-v0.3] | 2024-05-22 | | | | |
| RedHatAI/Mistral-7B-Instruct-v0.3-FP8 | 2024-06-26 | 38.3K | | | |
| RedHatAI/Mistral-7B-Instruct-v0.3-GPTQ-4bit | 2024-05-23 | 1.4M | | | |
| RedHatAI/Mistral-7B-Instruct-v0.3-quantized.w4a16 | 2024-07-11 | 7.6K | | | |
| RedHatAI/Mistral-7B-Instruct-v0.3-quantized.w8a16 | 2024-07-10 | 32.2K | | | |
| RedHatAI/Mistral-7B-Instruct-v0.3-quantized.w8a8 | 2024-07-18 | 3.7K | | | |
| nvidia/Mistral-7B-Instruct-v0.3-ONNX-INT4 | 2024-11-15 | 0 | | | |
| ^ DELTA (earliest date, total DLs) | | | -176d | +1.5M | RH:5 NV:1 |
--- | --- | --- | --- | --- | --- | --- | ---
#!/usr/bin/env python3
"""Compare quantized models released by both RedHatAI and Nvidia on Hugging Face Hub."""
import os
import re
import sys
from datetime import datetime
import pandas as pd
DOWNLOADS_THRESHOLD = 100_000
CSV_PATH = "hf_quant_dataset.csv" # Created by hf_download_model.py
def fmt_downloads(n):
if n >= 1_000_000:
return f"{n / 1_000_000:.1f}M"
if n >= 1_000:
return f"{n / 1_000:.1f}K"
return str(n)
def delta_days(date_a, date_b):
"""Return signed days between two date strings (a - b). Positive means a is later."""
try:
da = datetime.strptime(date_a, "%Y-%m-%d")
db = datetime.strptime(date_b, "%Y-%m-%d")
d = (da - db).days
return f"{d:+d}d"
except (ValueError, TypeError):
return "N/A"
def delta_downloads(dl_a, dl_b):
"""Return formatted difference (a - b) with sign."""
d = dl_a - dl_b
sign = "+" if d > 0 else ""
return f"{sign}{fmt_downloads(d)}" if d >= 0 else f"-{fmt_downloads(abs(d))}"
def print_table(title, headers, rows):
print(f"\n{'=' * 60}")
print(f" {title}")
print(f"{'=' * 60}")
col_widths = [len(h) for h in headers]
for row in rows:
for i, cell in enumerate(row):
col_widths[i] = max(col_widths[i], len(str(cell)))
header_line = " | ".join(h.ljust(col_widths[i]) for i, h in enumerate(headers))
sep_line = "-|-".join("-" * col_widths[i] for i in range(len(headers)))
print(header_line)
print(sep_line)
for row in rows:
print(" | ".join(str(cell).ljust(col_widths[i]) for i, cell in enumerate(row)))
def load_from_csv():
"""Load data from CSV created by hf_download_model.py."""
if not os.path.exists(CSV_PATH):
print(f"ERROR: {CSV_PATH} not found.")
print(f"Please run hf_download_model.py first to fetch and cache the data.")
sys.exit(1)
df = pd.read_csv(CSV_PATH)
rh, nv = {}, {}
orig_cache = {}
for _, row in df.iterrows():
info = {
"downloads": row["downloads"],
"created": row["created"],
}
if row["hub"] == "RedHatAI":
rh[row["name"]] = info
elif row["hub"] == "nvidia":
nv[row["name"]] = info
# Build cache of original dates
if row["base_date"] != "FAILED":
orig_cache[row["norm_base"]] = row["base_date"]
return rh, nv, orig_cache
def main():
print(f"Loading data from {CSV_PATH}...")
rh, nv, orig_cache = load_from_csv()
print(f" RedHatAI quantized: {len(rh)}")
print(f" Nvidia quantized: {len(nv)}")
print(f" Download threshold: >{fmt_downloads(DOWNLOADS_THRESHOLD)} combined")
# Build normalized base groupings (needed for both tables)
rh_by_norm, nv_by_norm = {}, {}
for name in rh:
# Use norm_base from CSV by loading it again
pass
# Reload CSV to get norm_base info
df = pd.read_csv(CSV_PATH)
for _, row in df.iterrows():
if row["hub"] == "RedHatAI":
rh_by_norm.setdefault(row["norm_base"], []).append(row["name"])
elif row["hub"] == "nvidia":
nv_by_norm.setdefault(row["norm_base"], []).append(row["name"])
def get_orig_date(norm_base):
return orig_cache.get(norm_base, "N/A")
# --- Exact matches ---
exact_names = sorted(set(rh) & set(nv))
exact_rows = []
for name in exact_names:
r, n = rh[name], nv[name]
combined = r["downloads"] + n["downloads"]
if combined < DOWNLOADS_THRESHOLD:
continue
# Find norm_base from df
norm = df[df["name"] == name].iloc[0]["norm_base"]
exact_rows.append((
name,
get_orig_date(norm),
r["created"], fmt_downloads(r["downloads"]),
n["created"], fmt_downloads(n["downloads"]),
delta_days(r["created"], n["created"]),
delta_downloads(r["downloads"], n["downloads"]),
))
exact_rows.sort(key=lambda r: min(rh[r[0]]["created"], nv[r[0]]["created"]), reverse=True)
print_table(
f"EXACT MATCHES ({len(exact_rows)} models, >{fmt_downloads(DOWNLOADS_THRESHOLD)} combined) [delta = RedHatAI minus Nvidia]",
["Model Name", "Base Date", "RH Date", "RH DLs", "NV Date", "NV DLs", "Date Delta", "DL Delta"],
exact_rows,
)
# --- Fuzzy matches ---
def base_earliest_date(base):
all_dates = []
for n in rh_by_norm.get(base, []):
d = rh[n]["created"]
if d != "N/A":
all_dates.append(d)
for n in nv_by_norm.get(base, []):
d = nv[n]["created"]
if d != "N/A":
all_dates.append(d)
return min(all_dates) if all_dates else "9999"
common_bases_all = set(rh_by_norm) & set(nv_by_norm)
common_bases = sorted(common_bases_all, key=base_earliest_date, reverse=True)
fuzzy_rows = []
fuzzy_count = 0
for base in common_bases:
r_names = sorted(rh_by_norm[base])
n_names = sorted(nv_by_norm[base])
if all(n in exact_names for n in r_names) and all(n in exact_names for n in n_names):
continue
rh_total_dl = sum(rh[n]["downloads"] for n in r_names)
nv_total_dl = sum(nv[n]["downloads"] for n in n_names)
if rh_total_dl + nv_total_dl < DOWNLOADS_THRESHOLD:
continue
fuzzy_count += 1
orig_date = get_orig_date(base)
fuzzy_rows.append((base, f"[base: {base}]", orig_date, "", "", "", "", ""))
for rn in r_names:
r = rh[rn]
fuzzy_rows.append(("", f" RedHatAI/{rn}", r["created"], fmt_downloads(r["downloads"]), "", "", "", ""))
for nn in n_names:
n = nv[nn]
fuzzy_rows.append(("", f" nvidia/{nn}", n["created"], fmt_downloads(n["downloads"]), "", "", "", ""))
rh_dates = [rh[n]["created"] for n in r_names]
nv_dates = [nv[n]["created"] for n in n_names]
rh_earliest = min(d for d in rh_dates if d != "N/A") if any(d != "N/A" for d in rh_dates) else "N/A"
nv_earliest = min(d for d in nv_dates if d != "N/A") if any(d != "N/A" for d in nv_dates) else "N/A"
fuzzy_rows.append((
"", " ^ DELTA (earliest date, total DLs)", "", "",
delta_days(rh_earliest, nv_earliest),
delta_downloads(rh_total_dl, nv_total_dl),
f"RH:{len(r_names)} NV:{len(n_names)}",
"",
))
fuzzy_rows.append(("---", "---", "---", "---", "---", "---", "---", "---"))
print_table(
f"FUZZY MATCHES — same base model, different quant ({fuzzy_count} base models, >{fmt_downloads(DOWNLOADS_THRESHOLD)} combined) [delta = RH minus NV]",
["Base Model", "Full Name", "Date", "Downloads", "Date Delta", "DL Delta", "Variants", ""],
fuzzy_rows,
)
if __name__ == "__main__":
main()
hub name created downloads downloads_30d quant_format model_size_b norm_base base_date base_downloads
RedHatAI gemma-4-31B-it-FP8-block 2026-04-03 4358837 3797660 fp8 31.0 gemma-4-31b 2026-03-11 33431223.0
RedHatAI Meta-Llama-3.1-70B-Instruct-FP8 2024-07-23 1444439 45972 fp8 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
RedHatAI Sparse-Llama-3.1-8B-gsm8k-2of4-FP8-dynamic 2024-11-14 923 12 fp8 8.0 sparse-llama-3.1-8b-gsm8k-2of4 2024-11-05 576.0
RedHatAI Llama-3.3-70B-Instruct-FP8-dynamic 2024-12-11 2850696 51419 fp8 70.0 llama-3.3-70b-instruct 2024-11-26 12995295.0
RedHatAI Llama-3.3-70B-Instruct-quantized.w4a16 2025-01-02 97927 14249 w4a16 70.0 llama-3.3-70b-instruct 2024-11-26 12995295.0
RedHatAI OmniCoder-9B-FP8-Dynamic 2026-03-20 1195 122 fp8 9.0 omnicoder-9b 2026-03-12 53215.0
RedHatAI gemma-4-31B-it-NVFP4 2026-04-03 742673 353056 nvfp4 31.0 gemma-4-31b 2026-03-11 33431223.0
RedHatAI gemma-4-26B-A4B-it-NVFP4 2026-04-05 2564317 1119333 nvfp4 26.0 gemma-4-26b-a4b 2026-03-11 34957706.0
RedHatAI gemma-4-31B-it-FP8-dynamic 2026-04-06 792733 399942 fp8 31.0 gemma-4-31b 2026-03-11 33431223.0
RedHatAI Qwen3.6-35B-A3B-NVFP4 2026-04-17 6481659 1938768 nvfp4 35.0 qwen3.6-35b-a3b 2026-04-15 15661179.0
RedHatAI Qwen3.6-35B-A3B-FP8-dynamic 2026-04-27 27055 22586 fp8 35.0 qwen3.6-35b-a3b 2026-04-15 15661179.0
RedHatAI gemma-4-12B-it-NVFP4 2026-06-08 288995 288995 nvfp4 12.0 gemma-4-12b 2026-05-23 3395580.0
RedHatAI diffusiongemma-26B-A4B-it-FP8-dynamic 2026-06-09 284459 284459 fp8 26.0 diffusiongemma-26b-a4b 2026-06-09 1783259.0
RedHatAI Llama-2-7b-ultrachat200k-pruned_50-quantized-deepsparse 2024-03-15 278 3 deepsparse 7.0 llama-2-7b-ultrachat200k-pruned-50 2024-03-15 720.0
RedHatAI Llama-2-7b-ultrachat200k-pruned_70-quantized-deepsparse 2024-03-15 427 7 deepsparse 7.0 llama-2-7b-ultrachat200k-pruned-70 2024-03-15 1064.0
RedHatAI Llama-2-7b-evol-code-alpaca-pruned_50-quantized-deepsparse 2024-03-15 343 2 deepsparse 7.0 llama-2-7b-evol-code-alpaca-pruned-50 2024-03-15 389.0
RedHatAI Llama-2-7b-evol-code-alpaca-pruned_70-quantized-deepsparse 2024-03-15 324 3 deepsparse 7.0 llama-2-7b-evol-code-alpaca-pruned-70 2024-03-15 357.0
RedHatAI Llama-2-7b-dolphin-open_platypus-pruned_50-quantized-deepsparse 2024-03-15 278 2 deepsparse 7.0 llama-2-7b-dolphin-open-platypus-pruned-50 2024-03-15 353.0
RedHatAI Llama-2-7b-dolphin-open_platypus-pruned_70-quantized-deepsparse 2024-03-15 285 4 deepsparse 7.0 llama-2-7b-dolphin-open-platypus-pruned-70 2024-03-15 553.0
RedHatAI Meta-Llama-3-8B-Instruct-FP8 2024-04-25 188904 2605 fp8 8.0 llama-3-8b-instruct 2024-04-17 44161480.0
RedHatAI Mixtral-8x7B-Instruct-v0.1-AutoFP8 2024-04-25 31198 8 fp8 7.0 mixtral-8x7b-instruct-v0.1 2023-12-10 22498540.0
RedHatAI Llama-2-7b-cnn-daily-mail-pruned_50-quantized-deepsparse 2024-05-15 214 1 deepsparse 7.0 llama-2-7b-cnn-daily-mail-pruned-50 2024-05-15 172.0
RedHatAI Llama-2-7b-cnn-daily-mail-pruned_70-quantized-deepsparse 2024-05-17 182 2 deepsparse 7.0 llama-2-7b-cnn-daily-mail-pruned-70 2024-05-15 183.0
RedHatAI Meta-Llama-3-8B-Instruct-FP8-KV 2024-05-20 287325 31223 fp8 8.0 llama-3-8b-instruct 2024-04-17 44161480.0
RedHatAI Mistral-7B-Instruct-v0.3-GPTQ-4bit 2024-05-23 1388401 1290 gptq 7.0 mistral-7b-instruct-v0.3 2024-05-22 37068277.0
RedHatAI Meta-Llama-3-70B-Instruct-FP8 2024-05-24 456760 846 fp8 70.0 llama-3-70b-instruct 2024-04-17 6153278.0
RedHatAI Qwen2-72B-Instruct-FP8 2024-06-06 23210 1178 fp8 72.0 qwen2-72b-instruct 2024-05-28 1328438.0
RedHatAI Mixtral-8x22B-Instruct-v0.1-AutoFP8 2024-06-11 5053 3 fp8 22.0 mixtral-8x22b-instruct-v0.1 2024-04-16 6175042.0
RedHatAI Qwen2-0.5B-Instruct-FP8 2024-06-14 31144 1321 fp8 0.5 qwen2-0.5b-instruct 2024-06-03 7873956.0
RedHatAI Qwen2-1.5B-Instruct-FP8 2024-06-14 286747 46510 fp8 1.5 qwen2-1.5b-instruct 2024-06-03 27154329.0
RedHatAI Qwen2-7B-Instruct-FP8 2024-06-14 188406 665 fp8 7.0 qwen2-7b-instruct 2024-06-04 24954191.0
RedHatAI Meta-Llama-3-70B-Instruct-FP8-KV 2024-06-26 4529 4 fp8 70.0 llama-3-70b-instruct 2024-04-17 6153278.0
RedHatAI Mistral-7B-Instruct-v0.3-FP8 2024-06-26 38333 3810 fp8 7.0 mistral-7b-instruct-v0.3 2024-05-22 37068277.0
RedHatAI Llama-2-7b-chat-hf-FP8 2024-06-26 11243 47 fp8 7.0 llama-2-7b-chat-hf 2023-07-13 30380953.0
RedHatAI Llama-2-7b-chat-quantized.w8a16 2024-07-02 2589 7 w8a16 7.0 llama-2-7b-chat 2023-07-09 1341.0
RedHatAI Meta-Llama-3-8B-Instruct-quantized.w8a16 2024-07-02 118577 298 w8a16 8.0 llama-3-8b-instruct 2024-04-17 44161480.0
RedHatAI Meta-Llama-3-70B-Instruct-quantized.w8a16 2024-07-02 14463 18 w8a16 70.0 llama-3-70b-instruct 2024-04-17 6153278.0
RedHatAI Qwen2-0.5B-Instruct-quantized.w8a16 2024-07-02 1005 8 w8a16 0.5 qwen2-0.5b-instruct 2024-06-03 7873956.0
RedHatAI Qwen2-1.5B-Instruct-quantized.w8a16 2024-07-02 730 7 w8a16 1.5 qwen2-1.5b-instruct 2024-06-03 27154329.0
RedHatAI Qwen2-7B-Instruct-quantized.w8a16 2024-07-02 776 11 w8a16 7.0 qwen2-7b-instruct 2024-06-04 24954191.0
RedHatAI Qwen2-72B-Instruct-quantized.w8a16 2024-07-02 3624 5 w8a16 72.0 qwen2-72b-instruct 2024-05-28 1328438.0
RedHatAI gemma-2-9b-it-FP8 2024-07-08 50509 1630 fp8 9.0 gemma-2-9b 2024-06-24 9881731.0
RedHatAI Qwen2-0.5B-Instruct-quantized.w4a16 2024-07-09 4185 160 w4a16 0.5 qwen2-0.5b-instruct 2024-06-03 7873956.0
RedHatAI Qwen2-1.5B-Instruct-quantized.w4a16 2024-07-09 2416 18 w4a16 1.5 qwen2-1.5b-instruct 2024-06-03 27154329.0
RedHatAI Mistral-7B-Instruct-v0.3-quantized.w8a16 2024-07-10 32206 35 w8a16 7.0 mistral-7b-instruct-v0.3 2024-05-22 37068277.0
RedHatAI Llama-2-7b-chat-quantized.w8a8 2024-07-11 14378 17 w8a8 7.0 llama-2-7b-chat 2023-07-09 1341.0
RedHatAI Meta-Llama-3-8B-Instruct-quantized.w8a8 2024-07-11 83186 517 w8a8 8.0 llama-3-8b-instruct 2024-04-17 44161480.0
RedHatAI Qwen2-1.5B-Instruct-quantized.w8a8 2024-07-11 27468 664 w8a8 1.5 qwen2-1.5b-instruct 2024-06-03 27154329.0
RedHatAI Qwen2-72B-Instruct-quantized.w4a16 2024-07-11 4475 11 w4a16 72.0 qwen2-72b-instruct 2024-05-28 1328438.0
RedHatAI Qwen2-7B-Instruct-quantized.w4a16 2024-07-11 4617 76 w4a16 7.0 qwen2-7b-instruct 2024-06-04 24954191.0
RedHatAI Llama-2-7b-chat-quantized.w4a16 2024-07-11 1574 6 w4a16 7.0 llama-2-7b-chat 2023-07-09 1341.0
RedHatAI Meta-Llama-3-70B-Instruct-quantized.w4a16 2024-07-11 582577 22 w4a16 70.0 llama-3-70b-instruct 2024-04-17 6153278.0
RedHatAI Meta-Llama-3-8B-Instruct-quantized.w4a16 2024-07-11 6476 235 w4a16 8.0 llama-3-8b-instruct 2024-04-17 44161480.0
RedHatAI Mistral-7B-Instruct-v0.3-quantized.w4a16 2024-07-11 7567 429 w4a16 7.0 mistral-7b-instruct-v0.3 2024-05-22 37068277.0
RedHatAI Qwen2-0.5B-Instruct-quantized.w8a8 2024-07-12 5318 10 w8a8 0.5 qwen2-0.5b-instruct 2024-06-03 7873956.0
RedHatAI Qwen2-7B-Instruct-quantized.w8a8 2024-07-12 5553 11 w8a8 7.0 qwen2-7b-instruct 2024-06-04 24954191.0
RedHatAI Meta-Llama-3-70B-Instruct-quantized.w8a8 2024-07-14 850 29 w8a8 70.0 llama-3-70b-instruct 2024-04-17 6153278.0
RedHatAI Qwen2-72B-Instruct-quantized.w8a8 2024-07-15 2508 8 w8a8 72.0 qwen2-72b-instruct 2024-05-28 1328438.0
RedHatAI Qwen2-57B-A14B-Instruct-FP8 2024-07-16 17100 647 fp8 57.0 qwen2-57b-a14b-instruct 2024-06-04 291585.0
RedHatAI Mistral-7B-Instruct-v0.3-quantized.w8a8 2024-07-18 3721 311 w8a8 7.0 mistral-7b-instruct-v0.3 2024-05-22 37068277.0
RedHatAI Meta-Llama-3.1-8B-Instruct-FP8 2024-07-23 6268917 545513 fp8 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
RedHatAI Meta-Llama-3.1-8B-Instruct-FP8-dynamic 2024-07-23 728962 67934 fp8 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
RedHatAI Meta-Llama-3.1-70B-Instruct-FP8-dynamic 2024-07-23 95960 16377 fp8 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
RedHatAI Meta-Llama-3.1-405B-Instruct-FP8 2024-07-23 44518 295 fp8 405.0 llama-3.1-405b-instruct 2024-07-16 4205284.0
RedHatAI Meta-Llama-3.1-405B-Instruct-FP8-dynamic 2024-07-23 70336 2247 fp8 405.0 llama-3.1-405b-instruct 2024-07-16 4205284.0
RedHatAI Meta-Llama-3.1-8B-Instruct-quantized.w8a16 2024-07-24 94353 1898 w8a16 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
RedHatAI Meta-Llama-3.1-8B-Instruct-quantized.w8a8 2024-07-24 424040 37780 w8a8 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
RedHatAI Meta-Llama-3.1-70B-Instruct-quantized.w8a16 2024-07-24 14525 36 w8a16 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
RedHatAI Meta-Llama-3.1-8B-Instruct-quantized.w4a16 2024-07-26 1513783 67362 w4a16 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
RedHatAI Meta-Llama-3.1-70B-Instruct-quantized.w8a8 2024-07-29 225432 8859 w8a8 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
RedHatAI Meta-Llama-3.1-8B-FP8 2024-07-31 1373126 239165 fp8 8.0 llama-3.1-8b 2024-07-14 25602084.0
RedHatAI Meta-Llama-3.1-70B-FP8 2024-07-31 13046 240 fp8 70.0 llama-3.1-70b 2024-07-14 2220136.0
RedHatAI Meta-Llama-3.1-8B-quantized.w8a16 2024-07-31 37052 5 w8a16 8.0 llama-3.1-8b 2024-07-14 25602084.0
RedHatAI Meta-Llama-3.1-8B-quantized.w8a8 2024-07-31 19214 1312 w8a8 8.0 llama-3.1-8b 2024-07-14 25602084.0
RedHatAI Meta-Llama-3.1-70B-Instruct-quantized.w4a16 2024-07-31 1830651 125014 w4a16 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
RedHatAI starcoder2-15b-FP8 2024-07-31 8909 3 fp8 15.0 starcoder2-15b 2024-02-20 558380.0
RedHatAI starcoder2-7b-FP8 2024-07-31 441 7 fp8 7.0 starcoder2-7b 2024-02-20 819423.0
RedHatAI starcoder2-3b-FP8 2024-07-31 600 4 fp8 3.0 starcoder2-3b 2023-11-29 7197486.0
RedHatAI Meta-Llama-3.1-70B-quantized.w8a16 2024-08-01 26 6 w8a16 70.0 llama-3.1-70b 2024-07-14 2220136.0
RedHatAI Meta-Llama-3.1-405B-FP8 2024-08-01 1201 69 fp8 405.0 llama-3.1-405b 2024-07-16 20971859.0
RedHatAI Meta-Llama-3.1-70B-quantized.w8a8 2024-08-01 408 5 w8a8 70.0 llama-3.1-70b 2024-07-14 2220136.0
RedHatAI starcoder2-15b-quantized.w8a16 2024-08-01 1642 8 w8a16 15.0 starcoder2-15b 2024-02-20 558380.0
RedHatAI starcoder2-3b-quantized.w8a16 2024-08-01 397 9 w8a16 3.0 starcoder2-3b 2023-11-29 7197486.0
RedHatAI starcoder2-7b-quantized.w8a16 2024-08-01 439 9 w8a16 7.0 starcoder2-7b 2024-02-20 819423.0
RedHatAI starcoder2-3b-quantized.w8a8 2024-08-01 672 61 w8a8 3.0 starcoder2-3b 2023-11-29 7197486.0
RedHatAI starcoder2-7b-quantized.w8a8 2024-08-01 530 7 w8a8 7.0 starcoder2-7b 2024-02-20 819423.0
RedHatAI starcoder2-15b-quantized.w8a8 2024-08-01 453 12 w8a8 15.0 starcoder2-15b 2024-02-20 558380.0
RedHatAI gemma-2-2b-it-FP8 2024-08-07 16979 1080 fp8 2.0 gemma-2-2b 2024-07-16 10055650.0
RedHatAI Meta-Llama-3.1-405B-Instruct-quantized.w4a16 2024-08-09 23785 106 w4a16 405.0 llama-3.1-405b-instruct 2024-07-16 4205284.0
RedHatAI gemma-2-9b-it-quantized.w8a16 2024-08-13 16742 83 w8a16 9.0 gemma-2-9b 2024-06-24 9881731.0
RedHatAI gemma-2-2b-it-quantized.w8a16 2024-08-13 984 24 w8a16 2.0 gemma-2-2b 2024-07-16 10055650.0
RedHatAI gemma-2-2b-quantized.w8a16 2024-08-13 679 7 w8a16 2.0 gemma-2-2b 2024-07-16 10055650.0
RedHatAI gemma-2-9b-it-quantized.w4a16 2024-08-15 10309 264 w4a16 9.0 gemma-2-9b 2024-06-24 9881731.0
RedHatAI gemma-2-2b-it-quantized.w4a16 2024-08-15 19049 172 w4a16 2.0 gemma-2-2b 2024-07-16 10055650.0
RedHatAI SmolLM-1.7B-Instruct-quantized.w8a16 2024-08-16 2914 4 w8a16 1.7 smollm-1.7b-instruct 2024-07-15 344858.0
RedHatAI gemma-2-2b-it-quantized.w8a8 2024-08-16 7227 49 w8a8 2.0 gemma-2-2b 2024-07-16 10055650.0
RedHatAI gemma-2-9b-it-quantized.w8a8 2024-08-16 2568 161 w8a8 9.0 gemma-2-9b 2024-06-24 9881731.0
RedHatAI Meta-Llama-3.1-405B-Instruct-quantized.w8a8 2024-08-19 4157 17 w8a8 405.0 llama-3.1-405b-instruct 2024-07-16 4205284.0
RedHatAI Meta-Llama-3.1-405B-Instruct-quantized.w8a16 2024-08-19 2036 12 w8a16 405.0 llama-3.1-405b-instruct 2024-07-16 4205284.0
RedHatAI gemma-2-27b-it-quantized.w8a16 2024-08-22 224 5 w8a16 27.0 gemma-2-27b 2024-06-24 5154404.0
RedHatAI SmolLM-1.7B-Instruct-quantized.w8a8 2024-08-23 432 2 w8a8 1.7 smollm-1.7b-instruct 2024-07-15 344858.0
RedHatAI Llama-3.2-1B-Instruct-FP8-dynamic 2024-09-25 11682743 1639556 fp8 1.0 llama-3.2-1b-instruct 2024-09-18 82420560.0
RedHatAI Llama-3.2-3B-Instruct-FP8-dynamic 2024-09-25 78893 824 fp8 3.0 llama-3.2-3b-instruct 2024-09-18 45159369.0
RedHatAI Llama-3.2-1B-Instruct-quantized.w8a8 2024-09-25 431906 43652 w8a8 1.0 llama-3.2-1b-instruct 2024-09-18 82420560.0
RedHatAI Llama-3.2-3B-Instruct-quantized.w8a8 2024-09-25 64556 3418 w8a8 3.0 llama-3.2-3b-instruct 2024-09-18 45159369.0
RedHatAI Llama-3.2-90B-Vision-Instruct-FP8-dynamic 2024-09-25 444695 1861 fp8 90.0 llama-3.2-90b-vision-instruct 2024-09-19 742732.0
RedHatAI Llama-3.2-1B-Instruct-FP8 2024-09-26 5714955 841598 fp8 1.0 llama-3.2-1b-instruct 2024-09-18 82420560.0
RedHatAI Llama-3.2-3B-Instruct-FP8 2024-09-26 339082 1095 fp8 3.0 llama-3.2-3b-instruct 2024-09-18 45159369.0
RedHatAI Llama-3.2-1B-quantized.w8a8 2024-09-26 360469 64487 w8a8 1.0 llama-3.2-1b 2024-09-18 55627999.0
RedHatAI Llama-3.2-3B-quantized.w8a8 2024-09-26 38295 4013 w8a8 3.0 llama-3.2-3b 2024-09-18 13619348.0
RedHatAI Llama-3.2-1B-FP8 2024-09-26 240993 34836 fp8 1.0 llama-3.2-1b 2024-09-18 55627999.0
RedHatAI Qwen2.5-0.5B-quantized.w8a8 2024-10-09 2608 586 w8a8 0.5 qwen2.5-0.5b 2024-09-15 24544068.0
RedHatAI Qwen2.5-0.5B-Instruct-quantized.w8a8 2024-10-09 8479 70 w8a8 0.5 qwen2.5-0.5b-instruct 2024-09-16 50430710.0
RedHatAI Qwen2.5-1.5B-quantized.w8a8 2024-10-09 6618774 929780 w8a8 1.5 qwen2.5-1.5b 2024-09-15 12501141.0
RedHatAI Qwen2.5-1.5B-Instruct-quantized.w8a8 2024-10-09 692 31 w8a8 1.5 qwen2.5-1.5b-instruct 2024-09-17 192429890.0
RedHatAI Qwen2.5-3B-Instruct-quantized.w8a8 2024-10-09 984 29 w8a8 3.0 qwen2.5-3b-instruct 2024-09-17 104024465.0
RedHatAI Qwen2.5-7B-Instruct-quantized.w8a8 2024-10-09 16071 774 w8a8 7.0 qwen2.5-7b-instruct 2024-09-16 153771461.0
RedHatAI Qwen2.5-7B-quantized.w8a8 2024-10-09 7965 29 w8a8 7.0 qwen2.5-7b 2024-09-15 19197266.0
RedHatAI Qwen2.5-32B-quantized.w8a8 2024-10-09 22 6 w8a8 32.0 qwen2.5-32b 2024-09-15 4145585.0
RedHatAI Qwen2.5-32B-Instruct-quantized.w8a8 2024-10-09 2571 187 w8a8 32.0 qwen2.5-32b-instruct 2024-09-17 25868299.0
RedHatAI Qwen2.5-72B-Instruct-quantized.w8a8 2024-10-09 1918 136 w8a8 72.0 qwen2.5-72b-instruct 2024-09-16 7726199.0
RedHatAI Qwen2.5-72B-quantized.w8a8 2024-10-09 374 4 w8a8 72.0 qwen2.5-72b 2024-09-15 660757.0
RedHatAI Qwen2.5-0.5B-quantized.w8a16 2024-10-09 2271 1548 w8a16 0.5 qwen2.5-0.5b 2024-09-15 24544068.0
RedHatAI Qwen2.5-0.5B-Instruct-quantized.w8a16 2024-10-09 86 13 w8a16 0.5 qwen2.5-0.5b-instruct 2024-09-16 50430710.0
RedHatAI Qwen2.5-1.5B-quantized.w8a16 2024-10-09 453 77 w8a16 1.5 qwen2.5-1.5b 2024-09-15 12501141.0
RedHatAI Qwen2.5-3B-quantized.w8a16 2024-10-09 397 229 w8a16 3.0 qwen2.5-3b 2024-09-15 6436856.0
RedHatAI Qwen2.5-7B-quantized.w8a16 2024-10-09 558 272 w8a16 7.0 qwen2.5-7b 2024-09-15 19197266.0
RedHatAI Qwen2.5-7B-Instruct-quantized.w8a16 2024-10-09 25486 5768 w8a16 7.0 qwen2.5-7b-instruct 2024-09-16 153771461.0
RedHatAI Qwen2.5-32B-Instruct-quantized.w8a16 2024-10-09 44 4 w8a16 32.0 qwen2.5-32b-instruct 2024-09-17 25868299.0
RedHatAI Qwen2.5-32B-quantized.w8a16 2024-10-09 605 120 w8a16 32.0 qwen2.5-32b 2024-09-15 4145585.0
RedHatAI Qwen2.5-72B-quantized.w8a16 2024-10-09 229 5 w8a16 72.0 qwen2.5-72b 2024-09-15 660757.0
RedHatAI pixtral-12b-FP8-dynamic 2024-10-10 265115 988 fp8 12.0 pixtral-12b 2024-09-11 105126.0
RedHatAI Llama-3.1-Nemotron-70B-Instruct-HF-FP8-dynamic 2024-10-16 172238 267 fp8 70.0 llama-3.1-nemotron-70b-instruct-hf 2024-10-12 1871913.0
RedHatAI Qwen2.5-3B-Instruct-quantized.w8a16 2024-10-17 117 4 w8a16 3.0 qwen2.5-3b-instruct 2024-09-17 104024465.0
RedHatAI Qwen2.5-1.5B-Instruct-quantized.w8a16 2024-10-17 328 8 w8a16 1.5 qwen2.5-1.5b-instruct 2024-09-17 192429890.0
RedHatAI Qwen2.5-72B-Instruct-quantized.w8a16 2024-10-17 102 4 w8a16 72.0 qwen2.5-72b-instruct 2024-09-16 7726199.0
RedHatAI Qwen2.5-72B-Instruct-quantized.w4a16 2024-10-17 1073 25 w4a16 72.0 qwen2.5-72b-instruct 2024-09-16 7726199.0
RedHatAI Qwen2.5-72B-quantized.w4a16 2024-10-17 18 3 w4a16 72.0 qwen2.5-72b 2024-09-15 660757.0
RedHatAI Qwen2.5-32B-quantized.w4a16 2024-10-18 21 1 w4a16 32.0 qwen2.5-32b 2024-09-15 4145585.0
RedHatAI Qwen2.5-32B-Instruct-quantized.w4a16 2024-10-18 595 103 w4a16 32.0 qwen2.5-32b-instruct 2024-09-17 25868299.0
RedHatAI Qwen2.5-7B-quantized.w4a16 2024-10-18 1242 7 w4a16 7.0 qwen2.5-7b 2024-09-15 19197266.0
RedHatAI granite-3.0-8b-instruct-GGUF 2024-10-31 108 3 gguf 8.0 granite-3.0-8b-instruct 2024-10-02 773522.0
RedHatAI granite-3.0-3b-a800m-instruct-GGUF 2024-11-01 103 3 gguf 3.0 granite-3.0-3b-a800m-instruct 2024-10-03 72689.0
RedHatAI granite-3.0-2b-instruct-GGUF 2024-11-01 135 2 gguf 2.0 granite-3.0-2b-instruct 2024-10-02 209079.0
RedHatAI granite-3.0-1b-a400m-instruct-GGUF 2024-11-01 138 5 gguf 1.0 granite-3.0-1b-a400m-instruct 2024-10-03 25435.0
RedHatAI Qwen2-VL-72B-Instruct-FP8-dynamic 2024-11-14 3710 45 fp8 72.0 qwen2-vl-72b-instruct 2024-09-17 1031300.0
RedHatAI Sparse-Llama-3.1-8B-evolcodealpaca-2of4-FP8-dynamic 2024-11-15 230 2 fp8 8.0 sparse-llama-3.1-8b-evolcodealpaca-2of4 2024-11-21 358.0
RedHatAI Sparse-Llama-3.1-8B-ultrachat_200k-2of4-FP8-dynamic 2024-11-19 318 5 fp8 8.0 sparse-llama-3.1-8b-ultrachat-200k-2of4 2024-11-21 446.0
RedHatAI Sparse-Llama-3.1-8B-ultrachat_200k-2of4-quantized.w4a16 2024-11-20 2838 7 w4a16 8.0 sparse-llama-3.1-8b-ultrachat-200k-2of4 2024-11-21 446.0
RedHatAI Sparse-Llama-3.1-8B-evolcodealpaca-2of4-quantized.w4a16 2024-11-21 280 4 w4a16 8.0 sparse-llama-3.1-8b-evolcodealpaca-2of4 2024-11-21 358.0
RedHatAI Sparse-Llama-3.1-8B-gsm8k-2of4-quantized.w4a16 2024-11-21 319 4 w4a16 8.0 sparse-llama-3.1-8b-gsm8k-2of4 2024-11-05 576.0
RedHatAI Qwen2.5-0.5B-FP8-dynamic 2024-11-27 2615 232 fp8 0.5 qwen2.5-0.5b 2024-09-15 24544068.0
RedHatAI Qwen2.5-0.5B-Instruct-FP8-dynamic 2024-11-27 316 110 fp8 0.5 qwen2.5-0.5b-instruct 2024-09-16 50430710.0
RedHatAI Qwen2.5-1.5B-FP8-dynamic 2024-11-27 4965 26 fp8 1.5 qwen2.5-1.5b 2024-09-15 12501141.0
RedHatAI Qwen2.5-1.5B-Instruct-FP8-dynamic 2024-11-27 167 22 fp8 1.5 qwen2.5-1.5b-instruct 2024-09-17 192429890.0
RedHatAI Qwen2.5-3B-FP8-dynamic 2024-11-27 641 98 fp8 3.0 qwen2.5-3b 2024-09-15 6436856.0
RedHatAI Qwen2.5-3B-Instruct-FP8-dynamic 2024-11-27 1742 103 fp8 3.0 qwen2.5-3b-instruct 2024-09-17 104024465.0
RedHatAI Qwen2.5-7B-FP8-dynamic 2024-11-27 3510 41 fp8 7.0 qwen2.5-7b 2024-09-15 19197266.0
RedHatAI Qwen2.5-7B-Instruct-FP8-dynamic 2024-11-27 57752 8060 fp8 7.0 qwen2.5-7b-instruct 2024-09-16 153771461.0
RedHatAI Qwen2.5-32B-Instruct-FP8-dynamic 2024-11-27 8089 1178 fp8 32.0 qwen2.5-32b-instruct 2024-09-17 25868299.0
RedHatAI Qwen2.5-32B-FP8-dynamic 2024-11-28 18 2 fp8 32.0 qwen2.5-32b 2024-09-15 4145585.0
RedHatAI Qwen2.5-72B-FP8-dynamic 2024-11-28 5734 40 fp8 72.0 qwen2.5-72b 2024-09-15 660757.0
RedHatAI Qwen2.5-Math-72B-FP8-dynamic 2024-11-28 15 2 fp8 72.0 qwen2.5-math-72b 2024-09-16 50378.0
RedHatAI Qwen2.5-Math-72B-Instruct-FP8-dynamic 2024-11-28 170 89 fp8 72.0 qwen2.5-math-72b-instruct 2024-09-16 134870.0
RedHatAI Qwen2.5-Coder-32B-FP8-dynamic 2024-11-28 276 3 fp8 32.0 qwen2.5-coder-32b 2024-11-08 363150.0
RedHatAI Qwen2.5-Coder-32B-Instruct-FP8-dynamic 2024-11-28 222444 14543 fp8 32.0 qwen2.5-coder-32b-instruct 2024-11-06 9512265.0
RedHatAI Qwen2.5-Coder-14B-Instruct-FP8-dynamic 2024-11-28 139973 1901 fp8 14.0 qwen2.5-coder-14b-instruct 2024-11-06 10953355.0
RedHatAI Qwen2.5-Coder-14B-FP8-dynamic 2024-11-28 73 3 fp8 14.0 qwen2.5-coder-14b 2024-11-08 309802.0
RedHatAI Qwen2.5-Math-7B-FP8-dynamic 2024-11-28 16 2 fp8 7.0 qwen2.5-math-7b 2024-09-16 3542980.0
RedHatAI Qwen2.5-Coder-7B-Instruct-FP8-dynamic 2024-11-28 5866 514 fp8 7.0 qwen2.5-coder-7b-instruct 2024-09-17 17413961.0
RedHatAI Qwen2.5-Coder-7B-FP8-dynamic 2024-11-28 376 102 fp8 7.0 qwen2.5-coder-7b 2024-09-16 3970148.0
RedHatAI Qwen2.5-72B-Instruct-FP8-dynamic 2024-12-02 58713 37675 fp8 72.0 qwen2.5-72b-instruct 2024-09-16 7726199.0
RedHatAI Qwen2.5-14B-Instruct-FP8-dynamic 2024-12-02 241833 32794 fp8 14.0 qwen2.5-14b-instruct 2024-09-16 41315030.0
RedHatAI Qwen2.5-14B-FP8-dynamic 2024-12-02 2808 129 fp8 14.0 qwen2.5-14b 2024-09-15 3784685.0
RedHatAI Qwen2.5-14B-quantized.w8a8 2024-12-03 529 3 w8a8 14.0 qwen2.5-14b 2024-09-15 3784685.0
RedHatAI Qwen2.5-3B-quantized.w8a8 2024-12-03 236 4 w8a8 3.0 qwen2.5-3b 2024-09-15 6436856.0
RedHatAI Qwen2.5-14B-Instruct-quantized.w8a8 2024-12-10 6125 185 w8a8 14.0 qwen2.5-14b-instruct 2024-09-16 41315030.0
RedHatAI Qwen2.5-Math-7B-Instruct-FP8-dynamic 2024-12-11 83 5 fp8 7.0 qwen2.5-math-7b-instruct 2024-09-19 1780214.0
RedHatAI Qwen2.5-0.5B-quantized.w4a16 2024-12-17 2158 84 w4a16 0.5 qwen2.5-0.5b 2024-09-15 24544068.0
RedHatAI granite-3.1-8b-instruct-GGUF 2024-12-17 209 6 gguf 8.0 granite-3.1-8b-instruct 2024-12-06 1518844.0
RedHatAI Qwen2.5-1.5B-quantized.w4a16 2024-12-17 844 17 w4a16 1.5 qwen2.5-1.5b 2024-09-15 12501141.0
RedHatAI Qwen2.5-3B-quantized.w4a16 2024-12-17 1065 34 w4a16 3.0 qwen2.5-3b 2024-09-15 6436856.0
RedHatAI Mixtral-8x22B-v0.1-quantized.w4a16 2025-01-03 255 1 w4a16 22.0 mixtral-8x22b-v0.1 2024-04-16 4854835.0
RedHatAI Mixtral-8x7B-v0.1-quantized.w4a16 2025-01-03 829 119 w4a16 7.0 mixtral-8x7b-v0.1 2023-12-01 8343512.0
RedHatAI QwQ-32B-Preview-FP8-dynamic 2025-01-03 26 3 fp8 32.0 qwq-32b-preview 2024-11-27 1432887.0
RedHatAI QwQ-32B-Preview-quantized.w4a16 2025-01-03 271 5 w4a16 32.0 qwq-32b-preview 2024-11-27 1432887.0
RedHatAI Llama-3.1-Nemotron-70B-Instruct-HF-quantized.w4a16 2025-01-03 192 13 w4a16 70.0 llama-3.1-nemotron-70b-instruct-hf 2024-10-12 1871913.0
RedHatAI Llama-3.1-Nemotron-70B-Instruct-HF-quantized.w8a8 2025-01-03 144 3 w8a8 70.0 llama-3.1-nemotron-70b-instruct-hf 2024-10-12 1871913.0
RedHatAI QwQ-32B-Preview-quantized.w8a8 2025-01-03 124 7 w8a8 32.0 qwq-32b-preview 2024-11-27 1432887.0
RedHatAI granite-3.1-8b-instruct-FP8-dynamic 2025-01-07 3292 355 fp8 8.0 granite-3.1-8b-instruct 2024-12-06 1518844.0
RedHatAI granite-3.1-2b-instruct-FP8-dynamic 2025-01-07 566 47 fp8 2.0 granite-3.1-2b-instruct 2024-12-06 299163.0
RedHatAI granite-3.1-8b-instruct-quantized.w4a16 2025-01-07 16110 1133 w4a16 8.0 granite-3.1-8b-instruct 2024-12-06 1518844.0
RedHatAI granite-3.1-8b-instruct-quantized.w8a8 2025-01-07 13889 206 w8a8 8.0 granite-3.1-8b-instruct 2024-12-06 1518844.0
RedHatAI granite-3.1-2b-instruct-quantized.w8a8 2025-01-07 6663 12 w8a8 2.0 granite-3.1-2b-instruct 2024-12-06 299163.0
RedHatAI granite-3.1-2b-instruct-quantized.w4a16 2025-01-07 192746 52587 w4a16 2.0 granite-3.1-2b-instruct 2024-12-06 299163.0
RedHatAI granite-3.1-2b-base-quantized.w8a8 2025-01-16 6518 9 w8a8 2.0 granite-3.1-2b-base 2024-12-06 105380.0
RedHatAI granite-3.1-2b-base-quantized.w4a16 2025-01-16 449 6 w4a16 2.0 granite-3.1-2b-base 2024-12-06 105380.0
RedHatAI granite-3.1-2b-base-FP8-dynamic 2025-01-16 273 12 fp8 2.0 granite-3.1-2b-base 2024-12-06 105380.0
RedHatAI granite-3.1-8b-base-quantized.w8a8 2025-01-16 5535 9 w8a8 8.0 granite-3.1-8b-base 2024-12-06 122056.0
RedHatAI granite-3.1-8b-base-quantized.w4a16 2025-01-16 818 36 w4a16 8.0 granite-3.1-8b-base 2024-12-06 122056.0
RedHatAI granite-3.1-8b-base-FP8-dynamic 2025-01-16 177 6 fp8 8.0 granite-3.1-8b-base 2024-12-06 122056.0
RedHatAI Llama-3.3-70B-Instruct-quantized.w8a8 2025-01-20 207192 9203 w8a8 70.0 llama-3.3-70b-instruct 2024-11-26 12995295.0
RedHatAI Qwen2-VL-72B-Instruct-quantized.w4a16 2025-01-23 878 13 w4a16 72.0 qwen2-vl-72b-instruct 2024-09-17 1031300.0
RedHatAI Mistral-Small-24B-Instruct-2501-FP8-dynamic 2025-01-30 922403 2828 fp8 24.0 mistral-small-24b-instruct-2501 2025-01-28 7148090.0
RedHatAI Mistral-Small-24B-Instruct-2501-quantized.w4a16 2025-01-31 13161 315 w4a16 24.0 mistral-small-24b-instruct-2501 2025-01-28 7148090.0
RedHatAI DeepSeek-R1-Distill-Llama-8B-quantized.w8a8 2025-02-01 263788 317 w8a8 8.0 deepseek-r1-distill-llama-8b 2025-01-20 18791366.0
RedHatAI DeepSeek-R1-Distill-Llama-8B-quantized.w4a16 2025-02-01 22393 368 w4a16 8.0 deepseek-r1-distill-llama-8b 2025-01-20 18791366.0
RedHatAI DeepSeek-R1-Distill-Qwen-1.5B-FP8-dynamic 2025-02-01 8617 449 fp8 1.5 deepseek-r1-distill-qwen-1.5b 2025-01-20 19747914.0
RedHatAI DeepSeek-R1-Distill-Qwen-7B-FP8-dynamic 2025-02-01 30391 1251 fp8 7.0 deepseek-r1-distill-qwen-7b 2025-01-20 14472661.0
RedHatAI DeepSeek-R1-Distill-Qwen-14B-FP8-dynamic 2025-02-01 36486 473 fp8 14.0 deepseek-r1-distill-qwen-14b 2025-01-20 7940414.0
RedHatAI DeepSeek-R1-Distill-Qwen-32B-FP8-dynamic 2025-02-01 57694 3675 fp8 32.0 deepseek-r1-distill-qwen-32b 2025-01-20 25901239.0
RedHatAI DeepSeek-R1-Distill-Llama-8B-FP8-dynamic 2025-02-01 13086 392 fp8 8.0 deepseek-r1-distill-llama-8b 2025-01-20 18791366.0
RedHatAI DeepSeek-R1-Distill-Llama-70B-FP8-dynamic 2025-02-01 490648 4846 fp8 70.0 deepseek-r1-distill-llama-70b 2025-01-20 5285395.0
RedHatAI Qwen2-VL-72B-Instruct-quantized.w8a8 2025-02-03 578 23 w8a8 72.0 qwen2-vl-72b-instruct 2024-09-17 1031300.0
RedHatAI DeepSeek-R1-Distill-Llama-70B-quantized.w8a8 2025-02-04 22856 223 w8a8 70.0 deepseek-r1-distill-llama-70b 2025-01-20 5285395.0
RedHatAI DeepSeek-R1-Distill-Qwen-7B-quantized.w4a16 2025-02-04 13417 165 w4a16 7.0 deepseek-r1-distill-qwen-7b 2025-01-20 14472661.0
RedHatAI DeepSeek-R1-Distill-Qwen-14B-quantized.w8a8 2025-02-04 33128 20 w8a8 14.0 deepseek-r1-distill-qwen-14b 2025-01-20 7940414.0
RedHatAI DeepSeek-R1-Distill-Qwen-14B-quantized.w4a16 2025-02-04 16998 286 w4a16 14.0 deepseek-r1-distill-qwen-14b 2025-01-20 7940414.0
RedHatAI DeepSeek-R1-Distill-Qwen-32B-quantized.w4a16 2025-02-05 47667 682 w4a16 32.0 deepseek-r1-distill-qwen-32b 2025-01-20 25901239.0
RedHatAI DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8 2025-02-05 33691 174 w8a8 32.0 deepseek-r1-distill-qwen-32b 2025-01-20 25901239.0
RedHatAI DeepSeek-R1-Distill-Qwen-7B-quantized.w8a8 2025-02-05 51240 293 w8a8 7.0 deepseek-r1-distill-qwen-7b 2025-01-20 14472661.0
RedHatAI DeepSeek-R1-Distill-Qwen-1.5B-quantized.w8a8 2025-02-06 50173 63 w8a8 1.5 deepseek-r1-distill-qwen-1.5b 2025-01-20 19747914.0
RedHatAI Qwen2.5-VL-3B-Instruct-FP8-dynamic 2025-02-06 241416 69851 fp8 3.0 qwen2.5-vl-3b-instruct 2025-01-26 88876421.0
RedHatAI Qwen2.5-VL-7B-Instruct-FP8-Dynamic 2025-02-06 338133 5390 fp8 7.0 qwen2.5-vl-7b-instruct 2025-01-26 79189674.0
RedHatAI Qwen2.5-VL-72B-Instruct-FP8-dynamic 2025-02-06 359823 5024 fp8 72.0 qwen2.5-vl-72b-instruct 2025-01-27 6744401.0
RedHatAI DeepSeek-R1-Distill-Llama-70B-quantized.w4a16 2025-02-07 49344 6656 w4a16 70.0 deepseek-r1-distill-llama-70b 2025-01-20 5285395.0
RedHatAI DeepSeek-R1-Distill-Qwen-1.5B-quantized.w4a16 2025-02-07 6856 80 w4a16 1.5 deepseek-r1-distill-qwen-1.5b 2025-01-20 19747914.0
RedHatAI DeepSeek-R1-Llama-70B-quantized.w8a8 2025-02-07 220 3 w8a8 70.0 deepseek-r1-llama-70b 2025-01-20 5285395.0
RedHatAI Qwen2.5-VL-7B-Instruct-quantized.w4a16 2025-02-07 44505 3717 w4a16 7.0 qwen2.5-vl-7b-instruct 2025-01-26 79189674.0
RedHatAI Qwen2.5-VL-3B-Instruct-quantized.w4a16 2025-02-07 129671 115971 w4a16 3.0 qwen2.5-vl-3b-instruct 2025-01-26 88876421.0
RedHatAI Qwen2.5-VL-7B-Instruct-quantized.w8a8 2025-02-07 100186 11189 w8a8 7.0 qwen2.5-vl-7b-instruct 2025-01-26 79189674.0
RedHatAI Qwen2.5-VL-3B-Instruct-quantized.w8a8 2025-02-07 378762 121846 w8a8 3.0 qwen2.5-vl-3b-instruct 2025-01-26 88876421.0
RedHatAI Qwen2.5-VL-72B-Instruct-quantized.w8a8 2025-02-07 13105 285 w8a8 72.0 qwen2.5-vl-72b-instruct 2025-01-27 6744401.0
RedHatAI Qwen2.5-VL-72B-Instruct-quantized.w4a16 2025-02-07 12390 372 w4a16 72.0 qwen2.5-vl-72b-instruct 2025-01-27 6744401.0
RedHatAI pixtral-12b-quantized.w4a16 2025-02-08 35177 191 w4a16 12.0 pixtral-12b 2024-09-11 105126.0
RedHatAI pixtral-12b-quantized.w8a8 2025-02-08 4813 7 w8a8 12.0 pixtral-12b 2024-09-11 105126.0
RedHatAI Mixtral-8x22B-Instruct-v0.1-FP8 2025-02-26 12500 7330 fp8 22.0 mixtral-8x22b-instruct-v0.1 2024-04-16 6175042.0
RedHatAI Mistral-Small-24B-Instruct-2501-quantized.w8a8 2025-03-03 126465 20069 w8a8 24.0 mistral-small-24b-instruct-2501 2025-01-28 7148090.0
RedHatAI QwQ-32B-FP8-dynamic 2025-03-05 24588 1035 fp8 32.0 qwq-32b 2024-11-27 1432887.0
RedHatAI Mixtral-8x7B-Instruct-v0.1-FP8 2025-03-06 154925 1432 fp8 7.0 mixtral-8x7b-instruct-v0.1 2023-12-10 22498540.0
RedHatAI QwQ-32B-quantized.w8a8 2025-03-11 824 23 w8a8 32.0 qwq-32b 2024-11-27 1432887.0
RedHatAI Mistral-Small-3.1-24B-Instruct-2503-FP8-dynamic 2025-03-27 393502 21492 fp8 24.0 mistral-small-3.1-24b-instruct-2503 2025-03-11 3534828.0
RedHatAI Llama-4-Scout-17B-16E-Instruct-FP8-dynamic 2025-04-10 213125 10381 fp8 17.0 llama-4-scout-17b-16e-instruct 2025-04-02 6860901.0
RedHatAI Mistral-Small-3.1-24B-Instruct-2503-quantized.w4a16 2025-04-15 457909 1972 w4a16 24.0 mistral-small-3.1-24b-instruct-2503 2025-03-11 3534828.0
RedHatAI Mistral-Small-3.1-24B-Instruct-2503-quantized.w8a8 2025-04-15 33251 1316 w8a8 24.0 mistral-small-3.1-24b-instruct-2503 2025-03-11 3534828.0
RedHatAI Qwen2.5-7B-Instruct-quantized.w4a16 2025-04-16 7909 385 w4a16 7.0 qwen2.5-7b-instruct 2024-09-16 153771461.0
RedHatAI Llama-4-Scout-17B-16E-Instruct-quantized.w4a16 2025-04-25 672385 5321 w4a16 17.0 llama-4-scout-17b-16e-instruct 2025-04-02 6860901.0
RedHatAI gemma-3-4b-it-FP8-dynamic 2025-04-28 34382 1839 fp8 4.0 gemma-3-4b 2025-02-20 22033988.0
RedHatAI gemma-3-12b-it-FP8-dynamic 2025-04-28 51688 3470 fp8 12.0 gemma-3-12b 2025-03-01 20342230.0
RedHatAI gemma-3-27b-it-FP8-dynamic 2025-04-28 1016734 529481 fp8 27.0 gemma-3-27b 2025-03-01 15588759.0
RedHatAI Qwen3-4B-quantized.w4a16 2025-05-02 101987 15162 w4a16 4.0 qwen3-4b 2025-04-27 66110120.0
RedHatAI Qwen3-32B-FP8-dynamic 2025-05-02 136755 1212 fp8 32.0 qwen3-32b 2025-04-27 41610959.0
RedHatAI Qwen3-0.6B-FP8-dynamic 2025-05-02 24890 470 fp8 0.6 qwen3-0.6b 2025-04-27 151103506.0
RedHatAI Qwen3-8B-FP8-dynamic 2025-05-02 277824 55601 fp8 8.0 qwen3-8b 2025-04-27 86991025.0
RedHatAI Qwen3-4B-FP8-dynamic 2025-05-02 11203 5694 fp8 4.0 qwen3-4b 2025-04-27 66110120.0
RedHatAI Qwen3-14B-FP8-dynamic 2025-05-02 11759 715 fp8 14.0 qwen3-14b 2025-04-27 22293281.0
RedHatAI Qwen3-1.7B-FP8-dynamic 2025-05-02 15568 1826 fp8 1.7 qwen3-1.7b 2025-04-27 48575630.0
RedHatAI Qwen3-1.7B-quantized.w4a16 2025-05-02 26456 7122 w4a16 1.7 qwen3-1.7b 2025-04-27 48575630.0
RedHatAI Qwen3-0.6B-quantized.w4a16 2025-05-02 30257 3479 w4a16 0.6 qwen3-0.6b 2025-04-27 151103506.0
RedHatAI Qwen3-30B-A3B-FP8-dynamic 2025-05-03 200203 5512 fp8 30.0 qwen3-30b-a3b 2025-04-27 14822289.0
RedHatAI Qwen3-235B-A22B-FP8-dynamic 2025-05-04 30975 176 fp8 235.0 qwen3-235b-a22b 2025-04-27 5732250.0
RedHatAI Qwen3-32B-quantized.w4a16 2025-05-05 51353 567 w4a16 32.0 qwen3-32b 2025-04-27 41610959.0
RedHatAI Qwen3-14B-quantized.w4a16 2025-05-05 11107 821 w4a16 14.0 qwen3-14b 2025-04-27 22293281.0
RedHatAI Qwen3-8B-quantized.w4a16 2025-05-06 161042 33685 w4a16 8.0 qwen3-8b 2025-04-27 86991025.0
RedHatAI Qwen3-30B-A3B-quantized.w4a16 2025-05-06 81590 39946 w4a16 30.0 qwen3-30b-a3b 2025-04-27 14822289.0
RedHatAI Llama-4-Maverick-17B-128E-Instruct-FP8 2025-05-12 58841 15136 fp8 17.0 llama-4-maverick-17b-128e-instruct 2025-04-01 676585.0
RedHatAI gemma-3-1b-it-FP8-dynamic 2025-05-16 7422 467 fp8 1.0 gemma-3-1b 2025-03-10 41661191.0
RedHatAI gemma-3-4b-it-quantized.w8a8 2025-06-04 5531 465 w8a8 4.0 gemma-3-4b 2025-02-20 22033988.0
RedHatAI gemma-3-4b-it-quantized.w4a16 2025-06-04 76554 6644 w4a16 4.0 gemma-3-4b 2025-02-20 22033988.0
RedHatAI gemma-3-12b-it-quantized.w8a8 2025-06-04 33222 1086 w8a8 12.0 gemma-3-12b 2025-03-01 20342230.0
RedHatAI gemma-3-12b-it-quantized.w4a16 2025-06-04 137393 38519 w4a16 12.0 gemma-3-12b 2025-03-01 20342230.0
RedHatAI gemma-3-27b-it-quantized.w8a8 2025-06-04 26211 772 w8a8 27.0 gemma-3-27b 2025-03-01 15588759.0
RedHatAI gemma-3-27b-it-quantized.w4a16 2025-06-04 947091 326056 w4a16 27.0 gemma-3-27b 2025-03-01 15588759.0
RedHatAI Sparse-Llama-3.1-8B-tldr-2of4-FP8-dynamic 2025-06-05 342 3 fp8 8.0 sparse-llama-3.1-8b-tldr-2of4 2025-05-16 60.0
RedHatAI gemma-3-1b-it-quantized.w8a8 2025-06-05 88321 296 w8a8 1.0 gemma-3-1b 2025-03-10 41661191.0
RedHatAI gemma-3-1b-it-quantized.w4a16 2025-06-05 10556 11 w4a16 1.0 gemma-3-1b 2025-03-10 41661191.0
RedHatAI Llama-3.1-8B-tldr-FP8-dynamic 2025-06-06 671 9 fp8 8.0 llama-3.1-8b-tldr 2024-09-28 693.0
RedHatAI Llama-4-Maverick-17B-128E-Instruct-quantized.w4a16 2025-06-12 23004 6439 w4a16 17.0 llama-4-maverick-17b-128e-instruct 2025-04-01 676585.0
RedHatAI Qwen3-30B-A3B-NVFP4 2025-06-17 190129 34470 nvfp4 30.0 qwen3-30b-a3b 2025-04-27 14822289.0
RedHatAI Mistral-Small-3.2-24B-Instruct-2506-FP8 2025-06-25 12880 354 fp8 24.0 mistral-small-3.2-24b-instruct-2506 2025-06-19 4750494.0
RedHatAI Llama-3.1-70B-Instruct-NVFP4 2025-06-27 5042 499 nvfp4 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
RedHatAI Llama-3.1-70B-Instruct-NVFP4A16 2025-06-27 183 28 nvfp4 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
RedHatAI Qwen3-32B-NVFP4 2025-06-27 139899 3864 nvfp4 32.0 qwen3-32b 2025-04-27 41610959.0
RedHatAI Qwen3-32B-NVFP4A16 2025-06-27 36506 284 nvfp4 32.0 qwen3-32b 2025-04-27 41610959.0
RedHatAI Qwen3-0.6B-FP8-BLOCK 2025-06-30 76383 16101 fp8 0.6 qwen3-0.6b 2025-04-27 151103506.0
RedHatAI Qwen3-30B-A3B-FP8-block 2025-07-01 96205 20028 fp8 30.0 qwen3-30b-a3b 2025-04-27 14822289.0
RedHatAI watt-tool-8B-FP8-dynamic 2025-07-02 21 6 fp8 8.0 watt-tool-8b 2024-12-19 735844.0
RedHatAI ToolACE-2-Llama-3.1-8B-FP8-dynamic 2025-07-02 16 3 fp8 8.0 toolace-2-llama-3.1-8b 2025-03-12 60721.0
RedHatAI Llama-4-Scout-17B-16E-Instruct-NVFP4 2025-07-08 20466 899 nvfp4 17.0 llama-4-scout-17b-16e-instruct 2025-04-02 6860901.0
RedHatAI SmolLM3-3B-FP8-dynamic 2025-07-28 6482 608 fp8 3.0 smollm3-3b 2025-07-08 3939411.0
RedHatAI SmolLM3-3B-quantized.w4a16 2025-07-31 946 28 w4a16 3.0 smollm3-3b 2025-07-08 3939411.0
RedHatAI gemma-3n-E4B-it-FP8-dynamic 2025-08-01 161958 464 fp8 4.0 gemma-3n-e4b 2025-06-03 1321114.0
RedHatAI gemma-3n-E2B-it-FP8-dynamic 2025-08-01 1760 118 fp8 2.0 gemma-3n-e2b 2025-06-12 3091542.0
RedHatAI gemma-3n-E4B-it-quantized.w8a8 2025-08-01 2246 19 w8a8 4.0 gemma-3n-e4b 2025-06-03 1321114.0
RedHatAI gemma-3n-E4B-it-quantized.w4a16 2025-08-01 1545 12 w4a16 4.0 gemma-3n-e4b 2025-06-03 1321114.0
RedHatAI gemma-3n-E2B-it-quantized.w8a8 2025-08-01 657 7 w8a8 2.0 gemma-3n-e2b 2025-06-12 3091542.0
RedHatAI gemma-3n-E2B-it-quantized.w4a16 2025-08-01 1200 9 w4a16 2.0 gemma-3n-e2b 2025-06-12 3091542.0
RedHatAI Llama-3.3-70B-Instruct-NVFP4 2025-08-05 26471 3201 nvfp4 70.0 llama-3.3-70b-instruct 2024-11-26 12995295.0
RedHatAI Voxtral-Mini-3B-2507-FP8-dynamic 2025-08-21 18187 654 fp8 3.0 voxtral-mini-3b-2507 2025-07-01 5114685.0
RedHatAI Qwen3-Coder-480B-A35B-Instruct-FP8 2025-09-04 1164 38 fp8 480.0 qwen3-coder-480b-a35b-instruct 2025-07-22 983122.0
RedHatAI Apertus-8B-Instruct-2509-FP8-dynamic 2025-09-18 159849 1481 fp8 8.0 apertus-8b-instruct-2509 2025-08-13 2544082.0
RedHatAI Apertus-70B-Instruct-2509-FP8-dynamic 2025-09-18 2329 648 fp8 70.0 apertus-70b-instruct-2509 2025-09-01 331604.0
RedHatAI Apertus-70B-Instruct-2509-quantized.w4a16 2025-09-21 178491 456 w4a16 70.0 apertus-70b-instruct-2509 2025-09-01 331604.0
RedHatAI Voxtral-Small-24B-2507-FP8-dynamic 2025-09-26 41857 7544 fp8 24.0 voxtral-small-24b-2507 2025-07-01 490950.0
RedHatAI Qwen3-VL-235B-A22B-Instruct-FP8-dynamic 2025-09-28 29624 31 fp8 235.0 qwen3-vl-235b-a22b-instruct 2025-09-22 6694922.0
RedHatAI Qwen3-VL-235B-A22B-Instruct-FP8-block 2025-09-28 1209 8 fp8 235.0 qwen3-vl-235b-a22b-instruct 2025-09-22 6694922.0
RedHatAI NVIDIA-Nemotron-Nano-9B-v2-FP8-dynamic 2025-09-30 33546 943 fp8 9.0 nvidia-nemotron-nano-9b-v2 2025-08-12 3323024.0
RedHatAI Llama-3.1-8B-Instruct-FP8-block 2025-10-10 724 7 fp8 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
RedHatAI Qwen3-14B-FP8-block 2025-10-12 262 12 fp8 14.0 qwen3-14b 2025-04-27 22293281.0
RedHatAI Qwen3-32B-FP8-block 2025-10-12 173 12 fp8 32.0 qwen3-32b 2025-04-27 41610959.0
RedHatAI Qwen3-8B-FP8-block 2025-10-12 936 37 fp8 8.0 qwen3-8b 2025-04-27 86991025.0
RedHatAI Qwen3-VL-235B-A22B-Instruct-NVFP4 2025-10-14 55941 6995 nvfp4 235.0 qwen3-vl-235b-a22b-instruct 2025-09-22 6694922.0
RedHatAI Llama-3.3-70B-Instruct-FP8-block 2025-10-20 25802 1777 fp8 70.0 llama-3.3-70b-instruct 2024-11-26 12995295.0
RedHatAI NVIDIA-Nemotron-Nano-9B-v2-quantized.w4a16 2025-10-22 8310 448 w4a16 9.0 nvidia-nemotron-nano-9b-v2 2025-08-12 3323024.0
RedHatAI Llama-4-Scout-17B-16E-Instruct-FP8-block 2025-10-23 724 5 fp8 17.0 llama-4-scout-17b-16e-instruct 2025-04-02 6860901.0
RedHatAI Llama-3.1-8B-Instruct-NVFP4 2025-10-23 128133 20212 nvfp4 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
RedHatAI Qwen3-14B-NVFP4 2025-10-23 75169 4034 nvfp4 14.0 qwen3-14b 2025-04-27 22293281.0
RedHatAI Qwen3-8B-NVFP4 2025-10-23 27837 3339 nvfp4 8.0 qwen3-8b 2025-04-27 86991025.0
RedHatAI Mistral-Small-3.2-24B-Instruct-2506-NVFP4 2025-10-23 91426 5047 nvfp4 24.0 mistral-small-3.2-24b-instruct-2506 2025-06-19 4750494.0
RedHatAI Llama-4-Maverick-17B-128E-Instruct-FP8-block 2025-10-27 139 7 fp8 17.0 llama-4-maverick-17b-128e-instruct 2025-04-01 676585.0
RedHatAI Llama-4-Maverick-17B-128E-Instruct-NVFP4 2025-10-27 10007 903 nvfp4 17.0 llama-4-maverick-17b-128e-instruct 2025-04-01 676585.0
RedHatAI Qwen3-235B-A22B-NVFP4 2025-10-29 1330 265 nvfp4 235.0 qwen3-235b-a22b 2025-04-27 5732250.0
RedHatAI Qwen3-235B-A22B-Instruct-2507-NVFP4 2025-10-29 12920 7067 nvfp4 235.0 qwen3-235b-a22b-instruct-2507 2025-07-21 1480496.0
RedHatAI Qwen3-4B-Thinking-2507-quantized.w4a16 2025-12-05 1223 39 w4a16 4.0 qwen3-4b-thinking-2507 2025-08-05 6455777.0
RedHatAI Qwen3-4B-Instruct-2507-quantized.w4a16 2025-12-05 1374 580 w4a16 4.0 qwen3-4b-instruct-2507 2025-08-05 52535540.0
RedHatAI Qwen3-30B-A3B-Thinking-2507-quantized.w4a16 2025-12-05 839 272 w4a16 30.0 qwen3-30b-a3b-thinking-2507 2025-07-29 3751954.0
RedHatAI Qwen3-30B-A3B-Instruct-2507-quantized.w4a16 2025-12-05 8164 1916 w4a16 30.0 qwen3-30b-a3b-instruct-2507 2025-07-28 13426532.0
RedHatAI Qwen3-VL-32B-Instruct-FP8-dynamic 2025-12-10 2404 651 fp8 32.0 qwen3-vl-32b-instruct 2025-10-19 13645955.0
RedHatAI Qwen3-VL-32B-Instruct-FP8-block 2025-12-10 108 3 fp8 32.0 qwen3-vl-32b-instruct 2025-10-19 13645955.0
RedHatAI Qwen3-VL-32B-Instruct-NVFP4 2025-12-10 38004 10720 nvfp4 32.0 qwen3-vl-32b-instruct 2025-10-19 13645955.0
RedHatAI Qwen3-Next-80B-A3B-Instruct-NVFP4 2025-12-12 79125 4767 nvfp4 80.0 qwen3-next-80b-a3b-instruct 2025-09-09 15688008.0
RedHatAI Qwen3-Next-80B-A3B-Instruct-FP8-dynamic 2025-12-12 214 8 fp8 80.0 qwen3-next-80b-a3b-instruct 2025-09-09 15688008.0
RedHatAI Qwen3-Next-80B-A3B-Instruct-FP8-block 2025-12-12 87 9 fp8 80.0 qwen3-next-80b-a3b-instruct 2025-09-09 15688008.0
RedHatAI Qwen3-Next-80B-A3B-Instruct-quantized.w4a16 2025-12-12 2886 399 w4a16 80.0 qwen3-next-80b-a3b-instruct 2025-09-09 15688008.0
RedHatAI Mistral-Large-3-675B-Instruct-2512-NVFP4 2025-12-23 100 6 nvfp4 675.0 mistral-large-3-675b-instruct-2512 2025-11-28 9123.0
RedHatAI NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 2025-12-23 9190 1061 fp8 30.0 nvidia-nemotron-3-nano-30b-a3b 2025-12-03 462372.0
RedHatAI Qwen3-Next-80B-A3B-Instruct-FP8 2025-12-24 5713 4719 fp8 80.0 qwen3-next-80b-a3b-instruct 2025-09-09 15688008.0
RedHatAI Qwen3-Next-80B-A3B-Thinking-quantized.w4a16 2025-12-24 572 42 w4a16 80.0 qwen3-next-80b-a3b-thinking 2025-09-09 2386070.0
RedHatAI Qwen3-Next-80B-A3B-Thinking-FP8-block 2025-12-24 138 9 fp8 80.0 qwen3-next-80b-a3b-thinking 2025-09-09 2386070.0
RedHatAI Qwen3-Next-80B-A3B-Thinking-FP8-dynamic 2025-12-24 1087 10 fp8 80.0 qwen3-next-80b-a3b-thinking 2025-09-09 2386070.0
RedHatAI Qwen3-Next-80B-A3B-Thinking-NVFP4 2025-12-24 1740 505 nvfp4 80.0 qwen3-next-80b-a3b-thinking 2025-09-09 2386070.0
RedHatAI Qwen3-4B-Instruct-2507-quantized.w8a8 2026-01-21 1830 310 w8a8 4.0 qwen3-4b-instruct-2507 2025-08-05 52535540.0
RedHatAI Qwen3-4B-Thinking-2507-quantized.w8a8 2026-01-21 342 17 w8a8 4.0 qwen3-4b-thinking-2507 2025-08-05 6455777.0
RedHatAI Qwen3-30B-A3B-Thinking-2507-quantized.w8a8 2026-01-21 537 453 w8a8 30.0 qwen3-30b-a3b-thinking-2507 2025-07-29 3751954.0
RedHatAI Qwen3-30B-A3B-Instruct-2507-quantized.w8a8 2026-01-21 4247 3207 w8a8 30.0 qwen3-30b-a3b-instruct-2507 2025-07-28 13426532.0
RedHatAI Ministral-3-14B-Instruct-2512-FP8-dynamic 2026-01-23 1749 241 fp8 14.0 ministral-3-14b-instruct-2512 2025-10-31 1593707.0
RedHatAI Ministral-3-14B-Instruct-2512-NVFP4 2026-02-04 1343 674 nvfp4 14.0 ministral-3-14b-instruct-2512 2025-10-31 1593707.0
RedHatAI Llama-Guard-4-12B-FP8-dynamic 2026-02-09 511 58 fp8 12.0 llama-guard-4-12b 2025-04-23 1181909.0
RedHatAI Llama-Guard-4-12B-quantized.w8a8 2026-02-09 198 14 w8a8 12.0 llama-guard-4-12b 2025-04-23 1181909.0
RedHatAI Llama-Guard-4-12B-quantized.w4a16 2026-02-16 83462 82836 w4a16 12.0 llama-guard-4-12b 2025-04-23 1181909.0
RedHatAI Qwen3.5-397B-A17B-FP8-dynamic 2026-02-18 11825 4681 fp8 397.0 qwen3.5-397b-a17b 2026-02-16 4628491.0
RedHatAI Qwen3-Next-80B-A3B-Instruct-quantized.w8a8 2026-03-03 455 333 w8a8 80.0 qwen3-next-80b-a3b-instruct 2025-09-09 15688008.0
RedHatAI Qwen3.5-122B-A10B-FP8-dynamic 2026-03-03 7561 1642 fp8 122.0 qwen3.5-122b-a10b 2026-02-24 3830314.0
RedHatAI Qwen3.5-35B-A3B-FP8-dynamic 2026-03-07 9248 2118 fp8 35.0 qwen3.5-35b-a3b 2026-02-24 13190350.0
RedHatAI sarvam-30b-FP8-dynamic 2026-03-09 135784 90938 fp8 30.0 sarvam-30b 2026-03-03 228432.0
RedHatAI sarvam-105b-FP8-dynamic 2026-03-09 985 27 fp8 105.0 sarvam-105b 2026-03-03 114315.0
RedHatAI Qwen3.5-122B-A10B-NVFP4 2026-03-17 569591 268849 nvfp4 122.0 qwen3.5-122b-a10b 2026-02-24 3830314.0
RedHatAI Qwen3-235B-A22B-Instruct-2507-quantized.w8a8 2026-03-18 242 36 w8a8 235.0 qwen3-235b-a22b-instruct-2507 2025-07-21 1480496.0
RedHatAI NVIDIA-Nemotron-3-Super-120B-A12B-FP8 2026-03-26 5146 502 fp8 120.0 nvidia-nemotron-3-super-120b-a12b 2026-03-06 9119.0
RedHatAI Mistral-Small-4-119B-2603-NVFP4 2026-03-27 154 15 nvfp4 119.0 mistral-small-4-119b-2603 2026-01-23 357383.0
RedHatAI NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 2026-03-27 11315 2049 nvfp4 120.0 nvidia-nemotron-3-super-120b-a12b 2026-03-06 9119.0
RedHatAI gemma-4-26B-A4B-it-FP8-Dynamic 2026-04-06 2714454 1999313 fp8 26.0 gemma-4-26b-a4b 2026-03-11 34957706.0
RedHatAI Qwen3.5-4B-quantized.w8a8 2026-04-15 21256 15558 w8a8 4.0 qwen3.5-4b 2026-02-27 27389353.0
RedHatAI Qwen3.5-4B-quantized.w4a16 2026-04-16 6737 5310 w4a16 4.0 qwen3.5-4b 2026-02-27 27389353.0
RedHatAI Qwen3.5-9B-quantized.w8a8 2026-04-16 18930 13380 w8a8 9.0 qwen3.5-9b 2026-02-27 32952516.0
RedHatAI Qwen3.5-9B-quantized.w4a16 2026-04-24 15325 6011 w4a16 9.0 qwen3.5-9b 2026-02-27 32952516.0
RedHatAI Qwen3.6-35B-A3B-FP8 2026-05-01 26054 13287 fp8 35.0 qwen3.6-35b-a3b 2026-04-15 15661179.0
RedHatAI Qwen3.6-27B-FP8 2026-05-01 54984 34378 fp8 27.0 qwen3.6-27b 2026-04-21 12695526.0
RedHatAI Qwen3.5-4B-FP8-dynamic 2026-05-11 29503 10601 fp8 4.0 qwen3.5-4b 2026-02-27 27389353.0
RedHatAI Qwen3.5-9B-FP8-dynamic 2026-05-11 313156 179418 fp8 9.0 qwen3.5-9b 2026-02-27 32952516.0
RedHatAI NVIDIA-Nemotron-3-Super-120B-A12B-FP8-Dflash 2026-06-01 138 129 fp8 120.0 nvidia-nemotron-3-super-120b-a12b 2026-03-06 9119.0
RedHatAI Mellum2-12B-A2.5B-Thinking-FP8-Dynamic 2026-06-02 37 37 fp8 12.0 mellum2-12b-a2.5b-thinking 2026-05-26 29935.0
RedHatAI Mellum2-12B-A2.5B-Thinking-NVFP4 2026-06-03 290 277 nvfp4 12.0 mellum2-12b-a2.5b-thinking 2026-05-26 29935.0
RedHatAI NVIDIA-Nemotron-3-Ultra-550B-A55B-FP8-dynamic 2026-06-04 14765 14467 fp8 550.0 nvidia-nemotron-3-ultra-550b-a55b 2026-05-26 1328.0
RedHatAI NVIDIA-Nemotron-3-Ultra-550B-A55B-FP8-block 2026-06-04 7881 7851 fp8 550.0 nvidia-nemotron-3-ultra-550b-a55b 2026-05-26 1328.0
RedHatAI NVIDIA-Nemotron-3-Ultra-550B-A55B-quantized.w4a16 2026-06-04 2444 1730 w4a16 550.0 nvidia-nemotron-3-ultra-550b-a55b 2026-05-26 1328.0
RedHatAI gemma-4-12B-it-FP8-Dynamic 2026-06-08 26193 26193 fp8 12.0 gemma-4-12b 2026-05-23 3395580.0
RedHatAI diffusiongemma-26B-A4B-it-NVFP4 2026-06-09 606150 606150 nvfp4 26.0 diffusiongemma-26b-a4b 2026-06-09 1783259.0
RedHatAI NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 2026-06-09 48 48 nvfp4 550.0 nvidia-nemotron-3-ultra-550b-a55b 2026-05-26 1328.0
nvidia Qwen3.6-27B-NVFP4 2026-06-22 538687 538687 nvfp4 27.0 qwen3.6-27b 2026-04-21 12695526.0
nvidia Qwen3.6-35B-A3B-NVFP4 2026-05-27 8492943 7114686 nvfp4 35.0 qwen3.6-35b-a3b 2026-04-15 15661179.0
nvidia Mistral-Medium-3.5-128B-NVFP4 2026-06-30 9273 9273 nvfp4 128.0 mistral-medium-3.5-128b 2026-03-31 767229.0
nvidia NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-NVFP4 2026-06-24 47 47 nvfp4 75.0 nvidia-nemotron-labs-3-puzzle-75b-a9b 2026-06-24 159.0
nvidia NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 2026-06-03 527337 462747 nvfp4 550.0 nvidia-nemotron-3-ultra-550b-a55b 2026-05-26 1328.0
nvidia NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 2026-03-10 5547346 1314206 nvfp4 120.0 nvidia-nemotron-3-super-120b-a12b 2026-03-06 9119.0
nvidia Gemma-4-26B-A4B-NVFP4 2026-05-01 3751276 2167115 nvfp4 26.0 gemma-4-26b-a4b 2026-03-11 34957706.0
nvidia Qwen3.5-122B-A10B-NVFP4 2026-05-13 93897 89426 nvfp4 122.0 qwen3.5-122b-a10b 2026-02-24 3830314.0
nvidia Gemma-4-31B-IT-NVFP4 2026-04-02 6634757 1644077 nvfp4 31.0 gemma-4-31b 2026-03-11 33431223.0
nvidia diffusiongemma-26B-A4B-it-NVFP4 2026-06-10 1687519 1687519 nvfp4 26.0 diffusiongemma-26b-a4b 2026-06-09 1783259.0
nvidia NVIDIA-Nemotron-3-Nano-4B-GGUF 2026-03-07 69030 17539 gguf 4.0 nvidia-nemotron-3-nano-4b 2025-05-03 186207.0
nvidia Qwen3.5-397B-A17B-NVFP4-V2 2026-06-29 19319 19319 nvfp4 397.0 qwen3.5-397b-a17b 2026-02-16 4628491.0
nvidia NVIDIA-Nemotron-3-Super-120B-A12B-FP8 2026-03-10 2232273 249979 fp8 120.0 nvidia-nemotron-3-super-120b-a12b 2026-03-06 9119.0
nvidia NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-FP8 2026-06-24 43 43 fp8 75.0 nvidia-nemotron-labs-3-puzzle-75b-a9b 2026-06-24 159.0
nvidia Llama-3.3-70B-Instruct-NVFP4 2025-01-16 588720 49272 nvfp4 70.0 llama-3.3-70b-instruct 2024-11-26 12995295.0
nvidia Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4 2026-04-24 3887330 2310420 nvfp4 30.0 nemotron-3-nano-omni-30b-a3b-reasoning 2026-04-20 1650159.0
nvidia Llama-3.1-8B-Instruct-NVFP4 2025-09-05 1031863 279780 nvfp4 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
nvidia Qwen3-Next-80B-A3B-Thinking-NVFP4 2025-12-11 210291 2868 nvfp4 80.0 qwen3-next-80b-a3b-thinking 2025-09-09 2386070.0
nvidia NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4 2025-12-20 3720048 1447033 nvfp4 30.0 nvidia-nemotron-3-nano-30b-a3b 2025-12-03 462372.0
nvidia Qwen3-Coder-480B-A35B-Instruct-NVFP4 2026-01-15 6334 992 nvfp4 480.0 qwen3-coder-480b-a35b-instruct 2025-07-22 983122.0
nvidia NVIDIA-Nemotron-3-Nano-4B-FP8 2026-03-12 96283 36474 fp8 4.0 nvidia-nemotron-3-nano-4b 2025-05-03 186207.0
nvidia Qwen3-VL-235B-A22B-Instruct-NVFP4-MLPerf-Inference-Closed-V6.1-FP8-KV 2026-06-15 18869 18869 fp8 235.0 qwen3-vl-235b-a22b-instruct 2025-09-22 6694922.0
nvidia Llama-3.1-8B-Instruct-FP8 2024-08-29 2313871 220917 fp8 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
nvidia Llama-3.1-70B-Instruct-FP8 2024-08-29 119577 3481 fp8 70.0 llama-3.1-70b-instruct 2024-07-16 23145855.0
nvidia Llama-3.1-405B-Instruct-FP8 2024-08-29 106076 4683 fp8 405.0 llama-3.1-405b-instruct 2024-07-16 4205284.0
nvidia Mistral-Nemo-12B-Instruct-ONNX-INT4 2024-11-13 0 0 int4 12.0 mistral-nemo-12b-instruct 2024-07-18 18989.0
nvidia Gemma-2b-it-ONNX-INT4 2024-11-14 0 0 int4 2.0 gemma-2b 2024-02-08 4753699.0
nvidia Meta-Llama-3.1-8B-Instruct-ONNX-INT4 2024-11-15 2031 12 int4 8.0 llama-3.1-8b-instruct 2024-07-18 167356088.0
nvidia Meta-Llama-3.2-3B-Instruct-ONNX-INT4 2024-11-15 0 0 int4 3.0 llama-3.2-3b-instruct 2024-09-18 45159369.0
nvidia Mistral-7B-Instruct-v0.3-ONNX-INT4 2024-11-15 0 0 int4 7.0 mistral-7b-instruct-v0.3 2024-05-22 37068277.0
nvidia Nemotron-Mini-4B-Instruct-ONNX-INT4 2024-11-15 0 0 int4 4.0 nemotron-mini-4b-instruct 2024-09-10 1091720.0
nvidia Llama-3.1-8B-Medusa-FP8 2025-01-13 1383 14 fp8 8.0 llama-3.1-8b-medusa 2025-02-08 40.0
nvidia Llama-3.1-405B-Instruct-NVFP4 2025-01-16 63233 1848 nvfp4 405.0 llama-3.1-405b-instruct 2024-07-16 4205284.0
nvidia Llama-4-Scout-17B-16E-Instruct-NVFP4 2025-04-14 584350 82819 nvfp4 17.0 llama-4-scout-17b-16e-instruct 2025-04-02 6860901.0
nvidia Llama-4-Maverick-17B-128E-Instruct-FP8 2025-04-14 15402 536 fp8 17.0 llama-4-maverick-17b-128e-instruct 2025-04-01 676585.0
nvidia Llama-4-Scout-17B-16E-Instruct-FP8 2025-04-14 1977775 337468 fp8 17.0 llama-4-scout-17b-16e-instruct 2025-04-02 6860901.0
nvidia Llama-3_1-Nemotron-Ultra-253B-v1-FP8 2025-04-30 31145 466 fp8 253.0 llama-3-1-nemotron-ultra-253b-v1 2025-04-07 611335.0
nvidia Llama-3.3-70B-Instruct-FP8 2025-05-05 652385 44380 fp8 70.0 llama-3.3-70b-instruct 2024-11-26 12995295.0
nvidia Llama-3_3-Nemotron-Super-49B-v1-FP8 2025-05-13 36617 1091 fp8 49.0 llama-3-3-nemotron-super-49b-v1 2025-03-16 995109.0
nvidia Nemotron-H-47B-Reasoning-128K-FP8 2025-05-29 5957 96 fp8 47.0 nemotron-h-47b-reasoning-128k 2025-05-22 12027.0
nvidia Nemotron-H-8B-Reasoning-128K-FP8 2025-06-05 3454 110 fp8 8.0 nemotron-h-8b-reasoning-128k 2025-06-05 115726.0
nvidia Qwen3-235B-A22B-FP8 2025-07-08 11453 204 fp8 235.0 qwen3-235b-a22b 2025-04-27 5732250.0
nvidia Qwen3-235B-A22B-NVFP4 2025-07-08 86858 6040 nvfp4 235.0 qwen3-235b-a22b 2025-04-27 5732250.0
nvidia Qwen3-30B-A3B-NVFP4 2025-07-08 705026 77705 nvfp4 30.0 qwen3-30b-a3b 2025-04-27 14822289.0
nvidia Llama-3_3-Nemotron-Super-49B-v1_5-FP8 2025-07-31 662762 236173 fp8 49.0 llama-3-3-nemotron-super-49b-v1-5 2025-07-25 1924492.0
nvidia Qwen3-8B-NVFP4 2025-09-09 444588 151863 nvfp4 8.0 qwen3-8b 2025-04-27 86991025.0
nvidia Qwen3-8B-FP8 2025-09-09 93143 4306 fp8 8.0 qwen3-8b 2025-04-27 86991025.0
nvidia Qwen3-14B-NVFP4 2025-09-09 549490 74501 nvfp4 14.0 qwen3-14b 2025-04-27 22293281.0
nvidia Qwen3-14B-FP8 2025-09-09 48112 6524 fp8 14.0 qwen3-14b 2025-04-27 22293281.0
nvidia Qwen3-32B-NVFP4 2025-09-09 627209 248135 nvfp4 32.0 qwen3-32b 2025-04-27 41610959.0
nvidia Qwen2.5-VL-7B-Instruct-FP8 2025-09-10 11667 982 fp8 7.0 qwen2.5-vl-7b-instruct 2025-01-26 79189674.0
nvidia Qwen2.5-VL-7B-Instruct-NVFP4 2025-09-10 392160 8092 nvfp4 7.0 qwen2.5-vl-7b-instruct 2025-01-26 79189674.0
nvidia NVIDIA-Nemotron-Nano-9B-v2-FP8 2025-09-22 873510 231553 fp8 9.0 nvidia-nemotron-nano-9b-v2 2025-08-12 3323024.0
nvidia Llama-3.1-Nemotron-Nano-VL-8B-V1-FP4-QAD 2025-10-01 4834 652 fp4-qad 8.0 llama-3.1-nemotron-nano-vl-8b-v1 2025-06-03 12958380.0
nvidia NVIDIA-Nemotron-Nano-9B-v2-NVFP4 2025-10-07 100660 19510 nvfp4 9.0 nvidia-nemotron-nano-9b-v2 2025-08-12 3323024.0
nvidia NVIDIA-Nemotron-Nano-12B-v2-VL-FP8 2025-10-22 1199421 28469 fp8 12.0 nvidia-nemotron-nano-12b-v2-vl 2025-10-21 1067576.0
nvidia NVIDIA-Nemotron-Nano-12B-v2-VL-NVFP4-QAD 2025-10-22 113897 27341 nvfp4 12.0 nvidia-nemotron-nano-12b-v2-vl 2025-10-21 1067576.0
nvidia Llama-3_3-Nemotron-Super-49B-v1_5-NVFP4 2025-11-11 49609 7144 nvfp4 49.0 llama-3-3-nemotron-super-49b-v1-5 2025-07-25 1924492.0
nvidia NVIDIA-Nemotron-3-Nano-30B-A3B-FP8 2025-12-06 5902087 349655 fp8 30.0 nvidia-nemotron-3-nano-30b-a3b 2025-12-03 462372.0
nvidia Qwen3-Next-80B-A3B-Instruct-NVFP4 2025-12-09 218134 46637 nvfp4 80.0 qwen3-next-80b-a3b-instruct 2025-09-09 15688008.0
nvidia Qwen3-235B-A22B-Thinking-2507-FP4-Eagle3 2025-12-15 260 51 fp4-eagle3 235.0 qwen3-235b-a22b-thinking-2507 2025-07-25 591612.0
nvidia Qwen3-VL-235B-A22B-Instruct-NVFP4 2025-12-25 16244 4228 nvfp4 235.0 qwen3-vl-235b-a22b-instruct 2025-09-22 6694922.0
nvidia Qwen3-235B-A22B-Thinking-2507-NVFP4 2025-12-30 27542 1535 nvfp4 235.0 qwen3-235b-a22b-thinking-2507 2025-07-25 591612.0
nvidia Qwen3-235B-A22B-Instruct-2507-NVFP4 2025-12-30 25296 3615 nvfp4 235.0 qwen3-235b-a22b-instruct-2507 2025-07-21 1480496.0
nvidia Qwen3-VL-235B-A22B-Instruct-NVFP4-MLPerf-Inference-Closed-V6.0 2026-01-27 190025 106913 unknown 235.0 qwen3-vl-235b-a22b-instruct 2025-09-22 6694922.0
nvidia Qwen3.5-397B-A17B-NVFP4 2026-02-16 2429467 480127 nvfp4 397.0 qwen3.5-397b-a17b 2026-02-16 4628491.0
nvidia NVIDIA-Nemotron-Labs-3-Elastic-30B-A3B-FP8 2026-04-01 4890 396 fp8 30.0 nvidia-nemotron-labs-3-elastic-30b-a3b 2026-04-01 5468.0
nvidia Qwen3-VL-235B-A22B-Instruct-NVFP4-MLPerf-Inference-Closed-V6.1 2026-04-07 6053 5989 unknown 235.0 qwen3-vl-235b-a22b-instruct 2025-09-22 6694922.0
nvidia NVIDIA-Nemotron-Labs-3-Elastic-30B-A3B-NVFP4 2026-04-14 6732 992 nvfp4 30.0 nvidia-nemotron-labs-3-elastic-30b-a3b 2026-04-01 5468.0
nvidia Nemotron-3-Nano-Omni-30B-A3B-Reasoning-FP8 2026-04-24 313431 111509 fp8 30.0 nemotron-3-nano-omni-30b-a3b-reasoning 2026-04-20 1650159.0
nvidia Wan2.2-T2V-A14B-Diffusers-NVFP4 2026-05-13 0 0 nvfp4 14.0 wan2.2-t2v-a14b-diffusers 2025-07-28 1588952.0
nvidia Wan2.2-T2V-A14B-Diffusers-FP8 2026-05-13 0 0 fp8 14.0 wan2.2-t2v-a14b-diffusers 2025-07-28 1588952.0
nvidia llama-nemotron-embed-vl-1b-v2-fp8 2026-05-14 6039 5948 fp8 1.0 llama-nemotron-embed-vl-1b-v2 2025-12-03 639215.0
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment