inference-optimization/ctest-Qwen3-8B-speculator.dspark-fp8ablation-bf16

ctest: Qwen3-8B DSpark speculator — FP8 hidden-states ablation (bf16)

Test checkpoint — not a production release. Trained to validate the FP8-quantized hidden-states transfer backend added in vllm-project/speculators#1028 (draft, reviving #491).

• Verifier: Qwen/Qwen3-8B
• Algorithm: dspark
• Hidden-states precision during training: bf16 (existing bf16 file backend)
• Training data: 5K magpie + 5K ultrachat samples from inference-optimization/Dataset-Qwen3-235B-Instruct

This is one of a matched pair (...-fp8ablation-bf16 / ...-fp8ablation-fp8) trained identically except for the hidden-states transfer precision, to isolate the effect of FP8 quantization on speculator quality. See the sibling repo for the other precision.

Validation metrics (final epoch)

• confidence_loss_epoch: 0.252827
• confidence_abs_error_epoch: 0.224563
• confidence_pred_mean_epoch: 0.499095
• confidence_cumprod_bias_epoch: 0.021723
• loss_epoch: 0.583228
• ce_loss_epoch: 1.192882
• tv_loss_epoch: 0.234570
• accept_rate_epoch: 0.455785
• accept_len_epoch: 2.873075
• full_acc_epoch: 0.482406
• position_0_acc_epoch: 0.698634
• position_1_acc_epoch: 0.592615
• position_2_acc_epoch: 0.525559
• position_3_acc_epoch: 0.472596
• position_4_acc_epoch: 0.433272
• position_5_acc_epoch: 0.400534
• position_6_acc_epoch: 0.376377
• position_7_acc_epoch: 0.358044

guidellm serving eval

See acceptance.csv in this repo for the full per-subset guidellm breakdown (9 subset rows).

Full ablation writeup, all 6 checkpoints' results, exact commands, and code: see the results package referenced from PR #1028.