Skip to content

Instantly share code, notes, and snippets.

View SALMANKHANPM's full-sized avatar
💭
Learning ....

P.M.SALMAN KHAN SALMANKHANPM

💭
Learning ....
View GitHub Profile
@SALMANKHANPM
SALMANKHANPM / train_math_nemo.py
Created February 6, 2026 04:11 — forked from shaltielshmid/train_math_nemo.py
NeMo-Framework training code from the paper "Learning to Reason: Training LLMs with GPT-OSS or DeepSeek R1 Reasoning Traces", for training Mistral-Nemo-Base-2407 or nvidia/NVIDIA-Nemotron-Nano-12B-v2-Base on reasoning data, generated either via gpt-oss-120b or DeepSeek-R1-0528
MODEL_TO_USE = "NanoV2" # or "MistralNemo"
IMPORT_MODEL_AND_DATA = False # set to True to import the model from HF hub, only needs to be run once
REASONING_STYLE = "gpt_oss_120b" # or DeepSeek_R1_0528
NUM_NODES = 2
GPUS_PER_NODE = 8
import nemo_run as run
from nemo.collections import llm
from nemo.collections.llm.gpt.model.mistral import MistralModel, MistralNeMoConfig12B
from nemo.collections.llm.gpt.model.ssm import MambaModel, NemotronNano12Bv2
@SALMANKHANPM
SALMANKHANPM / metalava_workaround_android-10
Created August 7, 2020 19:53 — forked from mikecriggs/metalava_workaround_android-10
Metalava issue workaround Android 10
If you're building Q and getting metalava errors apply:
https://github.com/Magma-WIP/build_soong/commit/bcd1bb529132905cf55e72f5a2a6ba19a99f60ac
https://github.com/Magma-WIP/build_soong/commit/dc3365fbde3b2a5773e655f690bb073967100795
If build fails even with the commits applied try building SystemUI with -j1
It can also be a good idea to increase swap
Command:
cd build/soong && git fetch https://github.com/Magma-WIP/build_soong ten-metalava && git cherry-pick bcd1bb529132905cf55e72f5a2a6ba19a99f60ac^..dc3365fbde3b2a5773e655f690bb073967100795