Safetensors

Latency-Sensitive Bench models

Inference-ready teachers and VLA policies for the supported benchmark tasks.

Layout

  • zero-latency/<task>/small-policy/ and zero-latency/<task>/vla/: models trained without latency.
  • latency-aware/<task>/small-policy/ and latency-aware/<task>/vla/: models trained for latency. Profile and fixed-2 training conditions are identified by run ID and bundle provenance.

MIKASA H8 conditioned inference bundles use latency-aware/mikasa-intercept-grab-fast/vla/starvla-<model>-h8/<run-id>/, where <model> is qwenoft, qwenpi_v3, or qwengr00t. Each model has an SFT run h8-conditioned-seed-reset-20260922-{profile|fixed-2} and a DAgger run h8-conditioned-dagger-20260923-{profile|fixed-2}.

Hist8 VLA release: 18 models, 27 formal evaluations

Flappy, Demon Attack and Deadly Corridor × QwenOFT, QwenGR00T and QwenPI v3 × zero/profile training. Each model is the evaluated step-5000 checkpoint, initialized from Qwen3-VL-4B revision ebb281ec70b05090aa6165b016eac8ec08e71b17 with seed 42.

Input is one current RGB image plus eight causal decision histories, passed as raw 0/1 transport state. Action horizon is 1; the model directory suffix h1 refers to this action horizon, while hist8 in each run ID refers to input history.

Formal scores

Each value is the mean of 100 episodes, seeds 1,000,000–1,000,099; parallel 32 and capacity 1. Flappy/Demon/Deadly caps are 3,600/7,200/3,600 raw frames at FPS/decision Hz 10/10, 60/15, 35/8.75. Deadly uses the corrected sf-render-v4 view (160×120 RGB with HUD).

Game / model zero→zero zero→profile profile→profile
flappy / qwenoft 439.13 19.19 373.78
flappy / qwengr00t 428.48 6.62 414.52
flappy / qwenpi_v3 408.72 5.58 344.48
demon_attack / qwenoft 2362.15 584.35 1452.80
demon_attack / qwengr00t 2370.90 231.90 1203.10
demon_attack / qwenpi_v3 2350.95 112.30 907.30
deadly_corridor / qwenoft 2098.67 742.35 2091.23
deadly_corridor / qwengr00t 2100.46 341.76 2090.82
deadly_corridor / qwenpi_v3 2104.09 140.94 1450.83

Downloadable models

Training Game / model Run
profile flappy / qwenoft flappy-qwenoft-hist8-scratch-s42-v2
profile demon_attack / qwenoft demon_attack-qwenoft-hist8-scratch-s42-v2
profile flappy / qwengr00t flappy-qwengr00t-hist8-scratch-s42-v2
profile demon_attack / qwengr00t demon_attack-qwengr00t-hist8-scratch-s42-v2
profile demon_attack / qwenpi_v3 demon_attack-qwenpi_v3-hist8-scratch-s42-v2-vlm3e-6
zero flappy / qwenoft flappy-qwenoft-hist8-zero-scratch-s42-v2
profile flappy / qwenpi_v3 flappy-qwenpi_v3-hist8-scratch-s42-v2-vlm3e-6
zero demon_attack / qwenoft demon_attack-qwenoft-hist8-zero-scratch-s42-v2
zero flappy / qwengr00t flappy-qwengr00t-hist8-zero-scratch-s42-v2
zero flappy / qwenpi_v3 flappy-qwenpi_v3-hist8-zero-scratch-s42-v2
zero deadly_corridor / qwenoft deadly_corridor-qwenoft-hist8-zero-scratch-s42-v2
zero demon_attack / qwengr00t demon_attack-qwengr00t-hist8-zero-scratch-s42-v2
profile deadly_corridor / qwenpi_v3 deadly_corridor-qwenpi_v3-hist8-scratch-s42-v2-vlm3e-6
profile deadly_corridor / qwenoft deadly_corridor-qwenoft-hist8-scratch-s42-v2
profile deadly_corridor / qwengr00t deadly_corridor-qwengr00t-hist8-scratch-s42-v2-workers16
zero demon_attack / qwenpi_v3 demon_attack-qwenpi_v3-hist8-zero-scratch-s42-v2
zero deadly_corridor / qwengr00t deadly_corridor-qwengr00t-hist8-zero-scratch-s42-v2
zero deadly_corridor / qwenpi_v3 deadly_corridor-qwenpi_v3-hist8-zero-scratch-s42-v2

Every model contains its README, checkpoint, inference/training configuration, statistics, task contract, collection/training provenance, exact latency profile and episode-level evaluation evidence. The original detailed JSON index remains in the benchmark publication receipt.

Interpretation and reproduction

The same zero-trained checkpoint is used in both evaluation environments. Each architecture has its own latency profile, so this is not a same-latency cross-architecture ranking. Zero and profile training use different teachers and collected data; differences do not isolate learning rate or a single training factor. Results use one training seed.

Common training: two GPUs, bf16 ZeRO-2, batch 64/rank, accumulation 1, global batch 128, 16 workers/rank, prefetch 4, backbone LR 3e-6, action-head LR 1e-4, 100-step warmup, cosine min_lr_rate 1/30, 5,000 steps. Architecture-specific losses and other parameter groups are retained.

Hist8 corrects the earlier -1/1 history input mismatch. Deadly also corrects an evaluation rendering mismatch; invalid older results are not included in this release. Profiles are preserved byte-for-byte with their sampling sidecars. Use each model task contract and pinned base-model revision when replaying.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support