Downloads · 30 days
6
18% of all-time downloads
MANGSEOK123/qwen3-4b-tau2-telecom_1_setting4_n91-1ep
qwen3-4b-tau2-telecom_1_setting4_n91-1ep is a machine learning model from MANGSEOK123. Use it for the machine 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.
Qwen3-4B-Instruct-2507 consolidated with OEL on the tau2-bench telecom domain, using 90 task–memory pairs.
Downloads · 30 days
6
18% of all-time downloads
All-time downloads
34
Public
Parameters
4.4B
8.8 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors8.8 GB · 100%
From the Hugging Face model README
Qwen3-4B-Instruct-2507 consolidated with OEL on the tau2-bench telecom domain, using 90 task–memory pairs.
| base model | Qwen3-4B-Instruct-2507 |
| training pairs | 90 |
| batch size | 6 |
| epochs | 1 |
| learning rate | 3e-6, constant |
| gradient clipping | 1.0 (verl default) |
| loss | full KL over all response tokens, kl_topk 256 |
| user simulator | gpt-4.1-mini (temperature 0) |
The student replays each task without memory; the teacher is the same weights with that task's memory in the system prompt. Only the prompt differs, and no reward is used.
| step | KL loss | entropy | grad norm |
|---|---|---|---|
| 1 | 0.010 | 0.422 | 2.826 |
| 2 | 0.013 | 0.191 | 1.764 |
| 3 | 0.009 | 0.437 | 1.603 |
| 4 | 0.008 | 0.199 | 1.688 |
| 5 | 0.019 | 0.405 | 1.299 |
| 6 | 0.018 | 0.181 | 5.537 |
| 7 | 0.012 | 0.419 | 1.305 |
| 8 | 0.012 | 0.496 | 0.721 |
| 9 | 0.027 | 0.191 | 1.882 |
| 10 | 0.008 | 0.278 | 1.842 |
| 11 | 0.018 | 0.156 | 12.795 |
| 12 | 0.088 | 0.199 | 14.023 |
| 13 | 0.017 | 0.279 | 1.227 |
| 14 | 0.038 | 0.242 | 3.597 |
| 15 | 0.010 | 0.299 | 3.833 |
Each step reads a different batch, so the loss column tracks batch difficulty rather than convergence.
Not evaluated. Pushed straight after training. For reference the base model scores
avg 0.056 / pass@4 0.175 on the tau2-bench telecom test split.
vllm serve MANGSEOK123/qwen3-4b-tau2-telecom_1_setting4_n91-1ep \
--enable-auto-tool-choice --tool-call-parser hermes --max-model-len 40960