# Health Check Report
# Generated: 2026-07-30T23:06 UTC
# Model: Nanthasit/sakthai-coder-1.5b
# Tool: sakthai-agent health-eval (cron)

model: Nanthasit/sakthai-coder-1.5b
eval_timestamp: 2026-07-30T23:06:00Z

metrics:
  popularity:
    downloads: 93
    likes: 0
    last_modified: 2026-07-30T22:56:24.000Z
    days_since_last_update: 0
  metadata:
    pipeline_tag: text-generation
    library_name: transformers
    license: apache-2.0
    base_model: Qwen/Qwen2.5-Coder-1.5B-Instruct
    private: false
    tags:
      - text-generation
      - transformers
      - qwen
      - coder
      - sakthai
      - finetuned
      - tool-calling
      - function-calling
      - 1.5b
      - llama.cpp
    datasets:
      - Nanthasit/sakthai-combined-v6
      - Nanthasit/sakthai-combined-v7
      - Nanthasit/sakthai-irrelevance-supplement
    datasets_count: 3

files:
  total_siblings: 1563
  model_weight_files: 1
  model_weights:
    - file: qwen2.5-coder-1.5b-instruct-q4_k_m.gguf
      size_bytes: 1117320768
      size_gb: 1.041
  total_model_size_gb: 1.041
  contamination: true
  contamination_note: "1563 siblings detected. Repo contains a full .venv/ directory (1254 .py files, pip packages, .dist-info). Only 1 actual model weight file. Recommends cleanup: remove .venv/ from repo."

benchmarks:
  model_index_present: true
  results:
    - dataset: HumanEval
      task: text-generation
      metric: pass@1 (base model reference)
      value: 74.4
      verified: false
    - dataset: MBPP
      task: text-generation
      metric: pass@1 (base model reference)
      value: 71.2
      verified: false
    - dataset: MultiPL-E (Python)
      task: text-generation
      metric: pass@1 (base model reference)
      value: 65.3
      verified: false
    - dataset: SakThai Coding Suite (internal)
      task: text-generation
      metric: pass@1 (fine-tuned model, internal)
      value: 100.0
      verified: false
      source: internal-local-llama-cpp-2026-07-25
  benchmark_quality:
    - "Internal benchmark (SakThai Coding Suite) shows 100% pass@1 but unverified — likely overfit or too narrow"
    - "Base model reference scores from Qwen2.5-Coder-1.5B-Instruct used as comparison; no verified fine-tuned scores"
    - "No multi-trial methodology reported; single-trial results may be unreliable"

health_score:
  overall: 6.5
  max: 10
  factors:
    popularity_downloads: 2
    likes: 0
    metadata_quality: 8
    model_index: 7
    file_hygiene: 3
    recency: 9
  concerns:
    - "Repo severely contaminated with .venv/ directory (1254 transient .py files)"
    - "0 likes — no community engagement yet"
    - "Benchmark scores unverified (marked verified: false)"
    - "Only 1 model weight format (GGUF Q4_K_M); no safetensors variant"
  recommendations:
    - action: REMOVE_VENV
      priority: high
      detail: "Delete .venv/ from repo — it adds 1254+ useless files and obscures real model content"
    - action: ADD_SAFETENSORS
      priority: medium
      detail: "Consider adding a safetensors version for Transformers-native loading"
    - action: VERIFY_BENCHMARKS
      priority: medium
      detail: "Run multi-trial benchmarks and mark as verified"
    - action: ADD_MODEL_CARD_SECTIONS
      priority: low
      detail: "Add usage example, recommended prompt format, and hardware requirements"