Li Wei
@li-wei-evalsRegular
Chinese-language model reviews, run on my own machines first.
- Model reviews
- Chinese models
- Agents
Shanghai, China · Chinese, English · Asia/Shanghai
- 86.3KBilibili @li-wei-evals followers
- 3.1KX @li-wei-evals followers
Pinned
Qwen3.5-35B-A3B 本地实测：值不值得换Nov 1, 2026VideoSource linked
- Establishes
- The link points to this creator's published work.
- Doesn’t establish
- Does not establish the numbers in it.
- Disclosure
- Independent work
Runs
16 verified · 7 on BenchClub cloud
ModelWhereSpeed
gemma-3-12b-it Q4_K_MNov 6, 2026RTX 4090 · 24 GB89.3 tok/s
MiniCPM5-2B Q4_K_MNov 6, 2026L40S · 48 GBCloud297.2 tok/s
Mistral-7B-Instruct-v0.2 Q4_K_MNov 2, 2026L40S · 48 GBCloud136.1 tok/s- gemma-3-27b-it Q4_K_MNov 1, 2026Apple M4 Max · 128 GB19.7 tok/s
Qwen3-8B Q4_K_MOct 29, 2026L40S · 48 GBCloud119.1 tok/s- Mistral-7B-Instruct-v0.2 Q4_K_MOct 29, 2026RTX 4090 · 24 GB156.2 tok/s
- Qwen3-14B Q4_K_MOct 24, 2026L40S · 48 GBCloud67.1 tok/s
Llama-3.1-8B-Instruct Q4_K_MOct 24, 2026Apple M4 Max · 128 GB67.8 tok/s
PassedFailedNot supported
Videos
GLM-4.7-Flash 工具调用实测Oct 18, 2026VideoSource linked
- Establishes
- The link points to this creator's published work.
- Doesn’t establish
- Does not establish the numbers in it.
- Disclosure
- Independent work
4090 还是 M4 Max：本地跑 MoE 怎么选Oct 10, 2026VideoSource linked
- Establishes
- The link points to this creator's published work.
- Doesn’t establish
- Does not establish the numbers in it.
- Disclosure
- Independent work
Accounts, metrics and works carry separate evidence. Updated Nov 6, 2026. Public JSON