prabakaranchandran/ex1-medbench-rl-medcalc-opd-lora-qwen3-0.6b
text-generation model by prabakaranchandran
Older snapshot: check the original source for current information.
From the publisher
Model card & documentation
Source previewRead the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.
Qwen3-0.6B-Base trained by on-policy distillation (OPD) on MedCalc-Bench Verified: the student writes its own answers, and the teacher (Qwen3-4B, post-trained) scores every token it wrote; each token's advantage is log πteacher − log πstudent (the per-token reverse KL). LoRA (r32, α32, all attention and MLP projections) merged into the dense weights. MedCalc-Bench Verified, full test (1,100 problems), greedy, vLLM. Each cell: greedy accuracy / pass@1 / pass@8 (%); pass@k on a 275-problem subset (5 per calculator), 8 samples at T = 1. Greedy accuracy on the full MedCalc-Bench Verified test (1,100 problems), condition C, at every exported checkpoint (vLLM). OPD (this model) and GRPO with the same LoRA configuration and step budget; horizontal lines: the base model and GRPO with full fine-tuning (58 steps). Final checkpoints: OPD 41.2% (161 steps), GRPO 32.4% (124 steps). Left: held-out accuracy during training (64 test problems × 8 samples at T…
Read the full model card ↗ · Preview checked 2026-10-30T15:30:55.960Z
Inside the original model card — Document outline
- ex1-medbench-rl · MedCalc-Bench: on-policy distillation from Qwen3-4B (Qwen3-0.6B-Base, LoRA) — EXPE
- Results
- Performance plots
- Training
- Limitations
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Documentation belongs to its respective authors. Reported project/model license: apache-2.0. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
Model overview
prabakaranchandran/ex1-medbench-rl-medcalc-opd-lora-qwen3-0.6b is a text-generation model repository published by prabakaranchandran on Hugging Face. The source reports the transformers library.
This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.
Model facts
- Task
- text-generation
- Library
- transformers
- Recent downloads (30 days)
- 0
- Cumulative likes
- 0
- Hugging Face trending score
- 0
- Reported safetensors parameters
- 596,049,920
- Architecture
- Qwen3ForCausalLM
- License
- apache-2.0
- Access
- Not gated by Hugging Face
- Created
- 2026-10-10T06:16:19.000Z
- Last modified
- 2026-10-10T06:17:13.000Z
Compatibility and lineage
Language coverage was not reported.
Reported base models: Qwen/Qwen3-0.6B-Base
Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.
Use and evaluate
Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.
No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.
Source and freshness
Source: Hugging Face Hub. Metadata observed 2026-10-10T06:41:01.047Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.