shant0602/openvla-7b-libero-spatial-lora-20k

OpenVLA-7B — LIBERO-Spatial LoRA (20k steps, with gradient clipping)

Fine-tuned from openvla/openvla-7b on the libero_spatial_no_noops task suite using LoRA (rank=32) for 20,000 gradient steps across two sequential runs on an NVIDIA GH200 (ARM64/aarch64).

Key contribution: includes a gradient clipping fix missing from the upstream OpenVLA training script — without it, training collapses at ~6k steps due to an unclamped gradient spike.

---

Evaluation Results

Evaluated on LIBERO-Spatial (10 tasks, 5 trials per task = 50 episodes total).

Setup	Steps	Score
Official paper (x86, flash-attn)	10k	84.7%
Official openvla-7b-finetuned-libero-spatial (ARM64 eval)	10k	37/50 = 74%
Our run 1 — no gradient clipping	10k	20/50 = 39%
Our run 2 — with gradient clipping	10k	28/50 = 56%
This checkpoint — grad clipping, 20k steps (ARM64 eval)	20k	36/50 = 72%

Our 20k checkpoint matches the official published 10k checkpoint (74% vs 72%, within statistical noise at 5 trials/task) on the same ARM64 hardware, confirming the gradient clipping fix is sufficient to recover training stability.

The ~12pp gap vs. the paper (84.7%) is a hardware effect: the official paper used x86 with native flash attention; on ARM64 the pre-built flash-attn binary is x86-only. Flash-attn 2.5.8 was built from source for run 3 but not run 2, so training used eager attention for most of the run. The official checkpoint evaluated at 74% on the same ARM64 setup, confirming this gap is hardware-driven, not a model quality issue.

Per-task breakdown (this checkpoint, 5 trials each)

Task	Successes	Rate
Task 1	3/5	60%
Task 2	4/5	80%
Task 3	4/5	80%
Task 4	5/5	100%
Task 5	3/5	60%
Task 6	2/5	40%
Task 7	5/5	100%
Task 8	5/5	100%
Task 9	3/5	60%
Task 10	2/5	40%
Total	36/50	72%

---

Training Curves (20k steps)

• Green = Run 2 (steps 0–10k, starting from base openvla/openvla-7b)
• Blue = Run 3 (steps 10k–20k, starting from the run 2 merged checkpoint)

Loss descends steadily from ~2.4 → ~0.68. Action token accuracy rises from ~0.05 → ~0.82. No loss spikes — contrast with run 1 (no gradient clipping) which spiked at step ~6k and never recovered.

---

Training Details

Setting	Value
Base model	openvla/openvla-7b
Dataset	libero_spatial_no_noops
Total steps	20,000 (2 × 10k sequential runs)
Batch size	16
Learning rate	5e-4
LoRA rank	32
LoRA dropout	0.0
Quantization	false
Image augmentation	true
Shuffle buffer	10,000
Hardware	NVIDIA GH200 (ARM64/aarch64, 480GB HBM3)

Resuming from a checkpoint

finetune.py has no native resume support. We resumed by setting VLA_PATH to the merged run 2 checkpoint. The model starts from those weights (all learned knowledge preserved); only the step counter and optimizer state reset. This is equivalent to a warm-start.

---

Key Fix: Gradient Clipping

The upstream vla-scripts/finetune.py in the OpenVLA repo is missing gradient norm clipping before optimizer.step(). A single batch with an outlier action can produce an unclamped gradient large enough to permanently corrupt the model weights.

What we saw: In run 1 (no clipping), loss jumped from 2.2 → 5.2 at step ~6k and action accuracy collapsed from 0.35 → 0.0 and never recovered. Final score: 39%.

The fix:

# Add before optimizer.step() in the training loop:
torch.nn.utils.clip_grad_norm_(trainable_params, max_norm=1.0)
optimizer.step()

With clipping, both runs (2 and 3) converged cleanly. See openvla/openvla#333 for the full analysis with loss curves.

---

Usage

This is a merged checkpoint — LoRA adapter weights are already merged into the base model. No PEFT library needed.

import torch
from transformers import AutoModelForVision2Seq, AutoProcessor

processor = AutoProcessor.from_pretrained(
    "shant0602/openvla-7b-libero-spatial-lora-20k",
    trust_remote_code=True,
)
model = AutoModelForVision2Seq.from_pretrained(
    "shant0602/openvla-7b-libero-spatial-lora-20k",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
).cuda()

# Inference
inputs = processor(image, instruction, return_tensors="pt").to("cuda")
action = model.predict_action(**inputs, unnorm_key="libero_spatial", do_sample=False)

The unnorm_key="libero_spatial" is required — it selects the action normalization statistics from dataset_statistics.json that were computed from the LIBERO-Spatial training data.