checkpoint_path string | configured_latency_ms float64 | eval_latency_steps int64 | experiment_name string | latency int64 | latency_type string | lengths list | mean_length float64 | mean_return float64 | returns list | seed int64 | std_return float64 | suite_name string | timestamp_utc string | train_latency_steps int64 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
results/Checkpoints/openvla_flappy_l2_full/final_model/pytorch_model.pt | 66.666667 | 2 | lora_cmp_flappy_l2_full | 2 | fixed_sweep | [
59,
69,
95,
435,
69,
59,
59,
365,
64,
104,
322,
64,
59,
512,
218,
59,
69,
64,
59,
286,
95,
219,
69,
59,
321,
64,
95,
59,
330,
276,
357,
64,
398,
132,
69,
59,
209,
59,
244,
59,
102,
399,
60,
60,
285,
69,
69,
174,
773,
6... | 167.56 | 18.518 | [
4.800000086426735,
6.70000009983778,
9.300000138580799,
51.400000631809235,
6.70000009983778,
4.800000086426735,
4.800000086426735,
43.50000052899122,
6.200000092387199,
11.100000150501728,
37.40000046789646,
6.200000092387199,
4.800000086426735,
60.900000743567944,
25.200000315904617,
... | 42 | 18.929889 | fixed_2 | 2026-06-02T23:20:12.557796+00:00 | 2 |
results/Checkpoints/openvla_flappy_l2_lora/final_model | 66.666667 | 2 | lora_cmp_flappy_l2_lora | 2 | fixed_sweep | [
50,
50,
31,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50,
50
] | 49.62 | 3.862 | [
3.90000007301569,
3.90000007301569,
2.0000000447034836,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.90000007301569,
3.900000073015... | 42 | 0.266 | fixed_2 | 2026-06-02T22:57:10.859128+00:00 | 2 |
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
LoRA vs Full Fine-Tuning — Flappy (fixed_l2) Ablation
Evaluation results for an ablation comparing LoRA vs full-parameter fine-tuning of
StarVLA/Qwen3-VL-4B-Instruct-Action (openvla framework) on the Flappy environment at a
fixed latency of 2 raw frames (66.67 ms).
Result (latency 2, 50 episodes, seed 42)
| Method | Mean reward | Std | Mean survival steps | Final eval CE loss | Model |
|---|---|---|---|---|---|
| Full FT | 18.52 | 18.93 | 167.6 | 0.060 | openvla_flappy_l2_full |
| LoRA (r=32, α=64) | 3.86 | 0.27 | 49.6 | 0.224 | openvla_flappy_l2_lora |
Full fine-tuning outperforms LoRA by ~4.8× in reward and survives ~3.4× longer. On this imitation task (CE on a Sample-Factory teacher), LoRA's low-rank update underfits relative to full FT, consistent with the higher held-out CE loss.
Controlled setup (everything identical except the variable under test)
- Data:
latency-sensitive-bench/flappy_fixed_l2_fs4(990 episodes, 948,645 frames), frame stack 4. - Loss: cross-entropy (categorical 2-action head). Steps: 2000. Seed: 42.
- Effective (mini) batch = 64 on both sides, tuned per GPU to fill memory:
- Full FT: per_device 32 × grad_accum 2 (single B200, ~151 GB peak).
- LoRA: per_device 16 × grad_accum 4 (single A100-80GB, ~73 GB peak; 32 OOMs).
- Learning rate (the only deliberately method-specific knob): VLM backbone LR 2e-5 (full) vs 2e-4 (LoRA, ~10×); the from-scratch action head is 1e-4 for both. Rationale: full FT updates all pretrained weights (small LR avoids catastrophic forgetting); LoRA's effective update is scaled by α/r and needs a ~10× larger LR — a shared LR would under-tune LoRA and bias the comparison.
- Eval:
latency_bench.run,fixedlatency = 2, 50 episodes. Both models were evaluated on the same A100 hardware for an identical eval environment (reward is deterministic given seed).
Files
full/queue_eval_latency_2.json,full/episode_metrics.jsonllora/queue_eval_latency_2.json,lora/episode_metrics.jsonl
- Downloads last month
- 15