exp-0216-005-db-balanced-qwen2.5-7b

Name: exp-0226-032-dynamic-template-hack-qwen2.5-7b
Rating: 2.5 (8 reviews)

Fine-tuned from Yano/exp-0212-001-alfworld-qwen2.5-7b (001 ALFWorld SFT model) using QLoRA (4-bit, Unsloth).

Purpose

DB Bench training with balanced data (v1-v4 mixed, INSERT/UPDATE downsampled). Addresses 004's SELECT degradation (76.5% -> 41.0%) caused by INSERT/UPDATE data imbalance.

Data Strategy

Combined all DB Bench v1-v4 (3,060 total)
Downsampled INSERT to 200, UPDATE to 150
Kept all SELECT/query-type samples
Final dataset: ~1470 samples (INSERT+UPDATE ~24.5%)

Training Configuration

Base model: Yano/exp-0212-001-alfworld-qwen2.5-7b
Method: QLoRA (4-bit), merged to 16-bit
Max sequence length: 2048
Epochs: 3
Learning rate: 2e-05
LoRA: r=64, alpha=128
Batch size: 2 (grad accum 16)
Warmup ratio: 0.1
Collator: AllAssistantTurnsCollator (all turns supervised)

Yano/exp-0226-032-dynamic-template-hack-qwen2.5-7b

exp-0216-005-db-balanced-qwen2.5-7b

Purpose

Data Strategy

Training Configuration

Yano/
exp-0226-032-dynamic-template-hack-qwen2.5-7b