06100df236
Passe de 95/3/2 (lookups/raisonnement/règles) à ~31/49/20. Dataset cible ~16K exemples denses (vs 66K de lookups avant). Modifiés : - 03_convert_cache.py : cache complet 1840 entrées (actuel + backup) - 04_build_dataset.py : subsampling agressif (CIM-10 1.5K, CCAM 1.5K, CoCoA 2K) + sélection intelligente priorisant le raisonnement - 12_generate_pipeline_examples.py : 3 templates (court + long + CPAM), cache actuel, cible ~2800 exemples Créés : - 13_generate_fascicule_reasoning.py : parsing 10 fascicules ATIH, génération Q&A raisonnement via Claude Opus 4.6 (~450 exemples) - 14_generate_negative_examples.py : 1000 exemples négatifs (symptômes/DP, redondances sémantiques, DAS non significatifs) - 15_generate_discrimination.py : 800 exercices de discrimination entre codes siblings CIM-10 via Claude Opus 4.6 - 16_parse_guide_metho.py : extraction Guide Méthodologique MCO 2026, Q&A directes + raisonnement via Claude Opus 4.6 (~500 exemples) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
68 lines
1.4 KiB
Bash
68 lines
1.4 KiB
Bash
#!/bin/bash
|
|
# Phase 2 — Installation d'Unsloth et dépendances pour le fine-tuning
|
|
#
|
|
# Prérequis vérifiés :
|
|
# - PyTorch 2.10.0+cu128 ✅
|
|
# - CUDA 12.x ✅
|
|
# - transformers 4.57.6 ✅
|
|
# - accelerate 1.12.0 ✅
|
|
#
|
|
# Usage : bash scripts/07_setup_unsloth.sh
|
|
|
|
set -e
|
|
|
|
VENV="/home/dom/ai/t2a/.venv"
|
|
PIP="$VENV/bin/pip"
|
|
PYTHON="$VENV/bin/python3"
|
|
|
|
echo "=== Installation Unsloth + dépendances ==="
|
|
echo ""
|
|
|
|
# 1. bitsandbytes (quantification 4-bit)
|
|
echo "[1/4] Installation de bitsandbytes..."
|
|
$PIP install --upgrade bitsandbytes
|
|
|
|
# 2. PEFT (LoRA)
|
|
echo ""
|
|
echo "[2/4] Installation de PEFT..."
|
|
$PIP install --upgrade peft
|
|
|
|
# 3. TRL (SFTTrainer)
|
|
echo ""
|
|
echo "[3/4] Installation de TRL..."
|
|
$PIP install --upgrade trl
|
|
|
|
# 4. Unsloth
|
|
echo ""
|
|
echo "[4/4] Installation d'Unsloth..."
|
|
$PIP install --upgrade --no-cache-dir "unsloth[cu128-torch2100] @ git+https://github.com/unslothai/unsloth.git"
|
|
|
|
# Vérification
|
|
echo ""
|
|
echo "=== Vérification ==="
|
|
$PYTHON -c "
|
|
import torch
|
|
print(f'PyTorch: {torch.__version__}')
|
|
print(f'CUDA: {torch.cuda.is_available()}')
|
|
print(f'GPU: {torch.cuda.get_device_name(0)}')
|
|
|
|
import bitsandbytes
|
|
print(f'bitsandbytes: OK')
|
|
|
|
from peft import LoraConfig
|
|
print(f'PEFT: OK')
|
|
|
|
from trl import SFTTrainer
|
|
print(f'TRL: OK')
|
|
|
|
try:
|
|
from unsloth import FastLanguageModel
|
|
print(f'Unsloth: OK')
|
|
except Exception as e:
|
|
print(f'Unsloth: {e}')
|
|
print('Essayez: pip install unsloth')
|
|
|
|
print()
|
|
print('=== Setup prêt pour le fine-tuning ! ===')
|
|
"
|