Model
- Model: deepseek-ai/DeepSeek-V4
- Total params: 1,600B
- Active params: 49B
- Architecture: MoE (ships with FP4/FP8 mixed-precision by default)
Why Cerebellum
DeepSeek-V4 already ships with its own mixed-precision scheme (FP4 for expert weights, FP8 elsewhere). But that's a uniform policy — Cerebellum could improve on it by testing which specific expert layers benefit from higher or lower precision. The model is ~860 GB even with default mixed-precision.
High community demand for better local deployment options.
What's needed
Way too large for consumer hardware ablation. Need access to serious iron (8x A100/H100 or equivalent).
If you have the hardware:
- Group ablation — test each tensor category at different precisions
- Layer ablation on groups that show variance
- Share PPL logs — we build the override file
Even partial data (just group-level ablation) would tell us if the method works at this scale.
Model
Why Cerebellum
DeepSeek-V4 already ships with its own mixed-precision scheme (FP4 for expert weights, FP8 elsewhere). But that's a uniform policy — Cerebellum could improve on it by testing which specific expert layers benefit from higher or lower precision. The model is ~860 GB even with default mixed-precision.
High community demand for better local deployment options.
What's needed
Way too large for consumer hardware ablation. Need access to serious iron (8x A100/H100 or equivalent).
If you have the hardware:
Even partial data (just group-level ablation) would tell us if the method works at this scale.