Downloads · 30 days
0
Zhongzhu/tunekv-bfcl-dsv4-flash
tunekv-bfcl-dsv4-flash is a machine learning model from Zhongzhu. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Tuned prefix KV for DeepSeek-V4-Flash-0731 on BFCL v4 (served via TuneKVConnectorDSV4, TUNEKVDSV4RETENTION=0).
Downloads · 30 days
0
Access
Public
Updated Sep 23, 2026
Repo size
509 MB
Likes
0
Public
Click a slice to open those files.
.safetensors256 MB · 84%
From the Hugging Face model README
Tuned prefix KV for DeepSeek-V4-Flash-0731 on BFCL v4 (served via TuneKVConnectorDSV4, TUNEKV_DSV4_RETENTION=0).
| arm | non-live (1390) | live (2251) | multi-turn (800) | total (4441) |
|---|---|---|---|---|
| base (same stack, 32K) | 1175 | 1649 | 489 | 3313 |
| rich system prompt | 1244 | 1835 | 554 | 3633 |
ce01/ CE + 0.1·KL | 1231 | 1831 | 549 | 3611 |
grpo/ GRPO + 0.1·KL | 1244 | 1839 | 542 | 3625 |
All tuned rows served with retention off (TUNEKV_DSV4_RETENTION=0). Eval: thinking/max, temp 1.0, 128K context, 32768-token per-step budget.