 hanzo-dev
|
a43c4c5f21
|
Add LICENSE and NOTICE files for license compliance
|
2026-06-27 01:43:54 +00:00 |
|
|
Hanzo Dev
|
2d05c5f5e8
|
chore: add LLM.md knowledge base, update .gitignore
|
2026-02-27 19:11:35 -08:00 |
|
|
z
|
3cde331205
|
Remove upstream credits
|
2026-02-26 19:56:00 -08:00 |
|
|
Zach Kelling
|
ea7e8832b6
|
fix: remove upstream model references, rebrand as Zen Max
Remove all third-party attribution. Zen Max is our model --
1.04T MoE, 32B active, 256K context. Keep benchmarks and specs.
|
2026-02-21 14:05:52 -08:00 |
|
|
Zach Kelling
|
105ac440e8
|
fix: correct params from 671B to 1.04T — Kimi K2-Thinking is 1,044B MoE
|
2026-02-21 11:58:43 -08:00 |
|
|
Zach Kelling
|
05c225aa30
|
chore: add Apache-2.0 LICENSE
|
2026-02-14 05:27:45 -08:00 |
|
|
Hanzo Dev
|
3989ff3f45
|
Add accurate hardware specs from Unsloth/Moonshot
- INT4 model: 370GB (62 shards)
- Minimum: 247GB combined RAM+VRAM+Disk
- GGUF quants: 245GB (1.66bit) to 588GB (4.5bit)
- Training: 4x A100 80GB needed (~500GB VRAM)
|
2025-11-17 00:38:56 -08:00 |
|
|
Hanzo Dev
|
523229ae69
|
Add note about base model weights location
|
2025-11-16 17:29:23 -08:00 |
|
|
Hanzo Dev
|
a03e17d772
|
Add model configuration and tokenizer files from Kimi K2 Thinking
- config.json: DeepseekV3 architecture with 384 experts, 8 active per token
- generation_config.json: Generation parameters
- tokenizer_config.json: Tokenizer configuration
- tiktoken.model: Tokenizer model (2.7MB)
Base model: moonshotai/Kimi-K2-Thinking
Architecture: DeepseekV3ForCausalLM (671B total, ~14B active)
|
2025-11-16 16:42:33 -08:00 |
|
|
Hanzo Dev
|
23f0fa86f2
|
Add YAML metadata and update architecture details (DeepseekV3 MoE 671B/14B active)
|
2025-11-16 16:38:18 -08:00 |
|
|
Hanzo Dev
|
77e3bd26f0
|
Add comprehensive README for zen-max based on Kimi K2 Thinking
|
2025-11-16 16:37:55 -08:00 |
|
|
Zach Kellingandsystem
|
ceb506060d
|
initial commit
|
2025-11-17 00:36:49 +00:00 |
|