GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
ZJUSCL
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
ZJUSCL
Model Tree
Qwen/Qwen3.5-9B
Input Modalities
Output Modalities
Supported Functionality
| Homepage | zjuscl.github.io/SeekJudge |
| Code | github.com/ZJUSCL/SeekJudge |
| Benchmark | ZJUSCL/CUAStepBench |
| Leaderboard | ZJUSCL/CUAStepBench_Leaderboard |
The repository README covers serving the model, running it as a batch judge
over trajectory datasets, and hosting it as an HTTP reward server for online RL
training. This checkpoint requires seek.trained: True in the SeekJudge
configuration, which selects the parser for the trained output format.
| Total parameters | 9.41 B (9,409.81 M) |
| Trainable during SFT | 8.95 B (8,953.80 M, 95.15 %) |
| Frozen | vision encoder and aligner (456 M) |
| Precision | bfloat16 |
| Checkpoint on disk | 18.8 GB (17.5 GiB), 4 safetensors shards |
| Context length | 262,144 (trained at 32,000) |
Architecture follows the Qwen3.5 multimodal stack: 32 language layers mixing linear attention with full attention every 4th layer, hidden size 4096, 16 attention heads with 4 KV heads, plus a 27-layer SigLIP-style vision tower (hidden size 1152, patch size 16) projecting to 4096.
| Hardware | 8 x NVIDIA RTX A6000 (48 GB), single node |
| Wall-clock time | 41.4 h (149,071 s) |
| GPU-hours | 331 A6000-hours |
| Throughput | 52.5 s per optimizer step, 0.61 samples/s |
| Peak memory | 25.4 GiB per GPU |
Peak memory stays low because DeepSpeed ZeRO-3 offloads both optimizer states and parameters to CPU, which trades memory for the step time above.
Full-parameter supervised fine-tuning of the language model on top of Qwen/Qwen3.5-9B, with the vision encoder and the aligner frozen. Run with ms-swift 4.1.0.dev0 on transformers 5.3.0.
| Parameter | Value |
|---|---|
| Tuning type | full (freeze_vit=True, freeze_aligner=True, freeze_llm=False) |
| Epochs | 2 |
| Optimizer steps | 2,840 (1,420 per epoch) |
| Global batch size | 32 sequences (1 per device x 4 accumulation x 8 GPUs) |
| Learning rate | 1e-5, cosine schedule, warmup ratio 0.05 |
| Optimizer | adamw_torch_fused, betas (0.9, 0.95), weight decay 0.01 |
| Samples | 45,416 |
| Tokens per sample | 3,335 mean, 2,971 std, 597 min, 31,769 max |
| Tokens per epoch | approximately 151 M |
| Tokens seen | approximately 303 M over 2 epochs |
The mixture combines trajectory-level judging targets, per-step analysis targets, and tool-query targets, so that one model serves every stage of the SeekJudge pipeline.
Training loss 0.246 and token accuracy 0.918 at step 2,840; mean loss over the whole run 0.380.
Apache-2.0, inherited from the Qwen3.5-9B base model. The SeekJudge codebase is MIT-licensed.
| Gradient clipping | 1.0 |
| Max sequence length | 32,000 |
| Precision | bfloat16 |
| Attention | FlashAttention |
| Memory | DeepSpeed ZeRO-3, CPU offload of optimizer and parameters, gradient checkpointing |
| Packing / padding-free | off |
| Template | qwen3_5, non-thinking prefix enabled |
| Seed | 42 |