Skip to content

Instantly share code, notes, and snippets.

View encredible's full-sized avatar
:atom:
moments of space

김재관 encredible

:atom:
moments of space
4 followers · 2 following
View GitHub Profile
@encredible
encredible / LOGICKOR_REPORT.md
Created August 16, 2026 04:17
Gaiel Korean LLM Benchmark (LogicKor Results)

Gaiel Series Korean Intelligence Benchmark (LogicKor)

Evaluated with GPT-4o Judge across 6 core Korean language & reasoning domains.

Overall Score Leaderboard

Model Single Turn Multi Turn Overall Score (10점 만점)
Gaiel-1.5B-Korean-Tuned-MLX 2.14 1.59 1.87
Gaiel-7B-Korean-Tuned-MLX 4.75 3.60 4.18
@encredible
encredible / LOGICKOR_REPORT.md
Created August 16, 2026 04:16
Gaiel Korean LLM Benchmark (LogicKor Results)

Gaiel Series Korean Intelligence Benchmark (LogicKor)

Evaluated with GPT-4o Judge across 6 core Korean language & reasoning domains.

Overall Score Leaderboard

Model Single Turn Multi Turn Overall Score (10점 만점)
Gaiel-1.5B-Korean-Tuned-MLX 2.15 1.60 1.87
Gaiel-7B-Korean-Tuned-MLX 4.75 3.60 4.18
@encredible
encredible / LOGICKOR_REPORT.md
Created August 16, 2026 03:20
Gaiel Korean LLM Benchmark (LogicKor Results)

Gaiel Series Korean Intelligence Benchmark (LogicKor)

Evaluated with GPT-4o Judge across 6 core Korean language & reasoning domains.

Overall Score Leaderboard

Model Single Turn Multi Turn Overall Score (10점 만점)
Gaiel-1.5B-Korean-Tuned-MLX 0.00 0.00 0.00
Gaiel-7B-Korean-Tuned-MLX 4.75 3.60 4.18
@encredible
encredible / LOGICKOR_REPORT.md
Created August 16, 2026 03:20
Gaiel Korean LLM Benchmark (LogicKor Results)

Gaiel Series Korean Intelligence Benchmark (LogicKor)

Evaluated with GPT-4o Judge across 6 core Korean language & reasoning domains.

Overall Score Leaderboard

Model Single Turn Multi Turn Overall Score (10점 만점)
Gaiel-1.5B-Korean-Tuned-MLX 0.00 0.00 0.00
Gaiel-7B-Korean-Tuned-MLX 4.75 3.60 4.18
@encredible
encredible / LOGICKOR_REPORT.md
Created August 16, 2026 03:14
Gaiel Korean LLM Benchmark (LogicKor Results)

Gaiel Series Korean Intelligence Benchmark (LogicKor)

Evaluated with GPT-4o Judge across 6 core Korean language & reasoning domains.

Overall Score Leaderboard

Model Single Turn Multi Turn Overall Score (10점 만점)
Gaiel-7B-Korean-Tuned-MLX 4.78 3.59 4.19
Gaiel-32B-Korean-Tuned-MLX 8.67 7.41 8.04
@encredible
encredible / benchmark_report.md
Created August 16, 2026 02:37
Gaiel MLX Benchmark Results (1.5B, 7B, 32B on Cluster)

Gaiel MLX Benchmark Results (Cluster: Mac mini + MacBook Air + MacBook Pro)

encredible/Gaiel-1.5B-Korean-Tuned-MLX

{
  "runs": [
    {
      "elapsed_s": 4.551236792001873,
      "output_text_preview": "A: \"a\" is a single character with a count of one. 0 characters are needed to represent this. \n\nB: \"a\" is the same as \"a a\". 0 characters are needed to represent this. \n\nC: \"abc\" is a three-character s",
@encredible
encredible / benchmark_report.md
Created August 16, 2026 02:37
Gaiel MLX Benchmark Results (1.5B, 7B, 32B on Cluster)

Gaiel MLX Benchmark Results (Cluster: Mac mini + MacBook Air + MacBook Pro)

encredible/Gaiel-1.5B-Korean-Tuned-MLX

encredible/Gaiel-7B-Korean-Tuned-MLX

encredible/Gaiel-32B-Korean-Tuned-MLX

encredible/Gaiel-72B-Korean-Tuned-MLX

Status: Model exceeds currently available cluster RAM limit for unquantized inference (Requires additional unified memory or 4-bit quantization).

@encredible
encredible / benchmark_report.md
Created August 16, 2026 02:25
Gaiel MLX Benchmark Results (3-Node Cluster)

Gaiel MLX Benchmark Results (Cluster: Mac mini + MacBook Air + MacBook Pro)

encredible/Gaiel-1.5B-Korean-Tuned-MLX

{
  "runs": [
    {
      "elapsed_s": 4.551236792001873,
      "output_text_preview": "A: \"a\" is a single character with a count of one. 0 characters are needed to represent this. \n\nB: \"a\" is the same as \"a a\". 0 characters are needed to represent this. \n\nC: \"abc\" is a three-character s",
      "stats": {
@encredible
encredible / benchmark_report.md
Created August 16, 2026 01:55
Gaiel MLX Benchmark Results

Gaiel MLX Benchmark Results

encredible/Gaiel-1.5B-Korean-Tuned-MLX

{
  "runs": [],
  "cluster": {
    "nodeIdentities": {
      "d3e74486ca588735ca37e0c78be109cd": {
        "modelId": "MacBook Pro",
@encredible
encredible / benchmark_report.md
Created August 16, 2026 01:48
Gaiel MLX Benchmark Results

Gaiel MLX Benchmark Results

encredible/Gaiel-1.5B-Korean-Tuned-MLX

Failed: Command failed with exit code 143

encredible/Gaiel-1.5B-Coding-Tuned-MLX

Failed: Command failed with exit code -15

encredible/Gaiel-7B-Korean-Tuned-MLX

Failed: Command failed with exit code 1

encredible/Gaiel-32B-Korean-Tuned-MLX

Failed: Command failed with exit code 1