Downloads · 30 days
18
8% of all-time downloads
deepseek-v2-lite-mobe-b4-bs16-h100
AverageMetaheuristicsEnjoyer/deepseek-v2-lite-mobe-b4-bs16-h100
deepseek-v2-lite-mobe-b4-bs16-h100 is a text generation model from AverageMetaheuristicsEnjoyer. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
mobe compression of deepseek-ai/DeepSeek-V2-Lite-Chat.
All-time downloads
237
Public
Parameters
13.3B
26.6 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors26.6 GB · 100%
At a glance
- Task
- Text Generation
- License
- apache-2.0
- Model type
- deepseek_v2
- Access
- Public
- Created
- May 31, 2026
- Updated
- May 31, 2026
- SHA
- 165863ae
Try a prompt
Base models
Model card
From the Hugging Face model README
mobe_deepseek_v2_lite
mobe compression of deepseek-ai/DeepSeek-V2-Lite-Chat.
- Method:
mobe - Base model: deepseek-ai/DeepSeek-V2-Lite-Chat
- Compressed:
gate_proj, up_projover layers 1-27 (down_projkept dense) - Settings: mobe, num_B=4, activation=tanh
See config.toml and training.json for the full run configuration and per-layer reconstruction stats.
Evaluation
| Task | Metric | Value |
|---|---|---|
| arc_challenge | sample_len | 1172.0000 |
| arc_challenge | acc,none | 0.4846 |
| arc_challenge | acc_stderr,none | 0.0146 |
| arc_challenge | acc_norm,none | 0.4949 |
| arc_challenge | acc_norm_stderr,none | 0.0146 |
| arc_easy | sample_len | 2376.0000 |
| arc_easy | acc,none | 0.7723 |
| arc_easy | acc_stderr,none | 0.0086 |
| arc_easy | acc_norm,none | 0.7593 |
| arc_easy | acc_norm_stderr,none | 0.0088 |
| winogrande | sample_len | 1267.0000 |
| winogrande | acc,none | 0.6906 |
| winogrande | acc_stderr,none | 0.0130 |
| piqa | sample_len | 1838.0000 |
| piqa | acc,none | 0.7982 |
| piqa | acc_stderr,none | 0.0094 |
| piqa | acc_norm,none | 0.8030 |
| piqa | acc_norm_stderr,none | 0.0093 |
| hellaswag | sample_len | 10042.0000 |
| hellaswag | acc,none | 0.5809 |
| hellaswag | acc_stderr,none | 0.0049 |
| hellaswag | acc_norm,none | 0.7626 |
| hellaswag | acc_norm_stderr,none | 0.0042 |
| openbookqa | sample_len | 500.0000 |
| openbookqa | acc,none | 0.3520 |
| openbookqa | acc_stderr,none | 0.0214 |
| openbookqa | acc_norm,none | 0.4500 |
| openbookqa | acc_norm_stderr,none | 0.0223 |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
"AverageMetaheuristicsEnjoyer/deepseek-v2-lite-mobe-b4-bs16-h100",
trust_remote_code=True,
torch_dtype="bfloat16",
device_map="auto",
)
tok = AutoTokenizer.from_pretrained("AverageMetaheuristicsEnjoyer/deepseek-v2-lite-mobe-b4-bs16-h100", trust_remote_code=True)
The factored experts are reconstructed at runtime via the bundled DeepseekV2MoBEForCausalLM.
- Task
- Text Generation
- Type
- deepseek_v2
- License
- apache-2.0
- Created
- May 31, 2026
- Updated
- May 31, 2026