Downloads · 30 days
10
26% of all-time downloads
VanWang/awm-qwen3.6-27b-rl-d1000-ep2
awm-qwen3.6-27b-rl-d1000-ep2 is a reinforcement learning model from VanWang. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Qwen3.6-27B trained with GRPO (Dr.GRPO variant) to act as an orchestrator in a two-agent system: it decomposes a task, delegates subtasks to a second model, and integrates the results. Only orchestrator tokens receive…
Downloads · 30 days
10
26% of all-time downloads
All-time downloads
38
Public
Parameters
27.4B
54.8 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors54.8 GB · 100%
From the Hugging Face model README
Qwen3.6-27B trained with GRPO (Dr.GRPO variant) to act as an orchestrator in a
two-agent system: it decomposes a task, delegates subtasks to a second model, and
integrates the results. Only orchestrator tokens receive gradient
(train_roles: ["orch"]) — the sub-agent is never trained.
| Base model | Qwen/Qwen3.6-27B |
| Algorithm | GRPO / Dr.GRPO (--disable-grpo-std-normalization) |
| Training set | 1001 tasks |
| Epochs | 2 |
| Samples per prompt | 8, DAPO dynamic sampling (over-sample 64 / batch 32) |
| Parallelism | 8× p5e nodes, tp×cp = 16, dp = 4 |
| Stack | slime + Megatron-LM + SGLang |
196 held-out tasks × 3 seeds = 588 runs per setting. All rows share identical
settings (n=3, base_seed=42, max_turns=200, orch_style=pure); the comparison
against the base model is a paired t-test over per-task means.
The two "fixed sub" rows hold the sub-agent constant, so any difference is attributable to orchestration alone rather than to the executor improving.
| setting | base Qwen3.6-27B | this model | paired diff |
|---|---|---|---|
| single agent (no delegation) | 0.3808 | 0.3924 | +0.0116 (t=+1.32) |
| self-paired (orch = sub = this model) | 0.3499 | 0.4034 | +0.0535 (t=+5.30) ✅ |
| fixed weak sub (Qwen3.6-35B-A3B) | 0.3110 | 0.3652 | +0.0542 (t=+4.99) ✅ |
| fixed strong sub (DeepSeek-V4-Flash) | 0.4153 | 0.4723 | +0.0571 (t=+5.20) ✅ |
Reference: DeepSeek-V4-Flash orchestrating itself scores 0.4726 in the strong-sub setting. The 1001-task checkpoints match that number exactly; the 498-task ones remain significantly below it.
sglang serve --model-path VanWang/awm-qwen3.6-27b-rl-d1000-ep2 --tp 8
The served model name must match what your client sends. The model expects the orchestration prompt format used during training.