Downloads · 30 days
9
23% of all-time downloads
albedo-qwen3.6-35b-bookend-phase4
JMaxCool/albedo-qwen3.6-35b-bookend-phase4
albedo-qwen3.6-35b-bookend-phase4 is a machine learning model from JMaxCool. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
LoRA DPO fine-tune on King CXXIV for native bookend emission (Phase 4).
All-time downloads
40
Public
Parameters
34.7B
71.9 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors71.9 GB · 100%
At a glance
- Model type
- qwen3_5_moe
- Access
- Public
- Created
- Sep 9, 2026
- Updated
- Sep 9, 2026
- SHA
- bd232e5c
Model card
From the Hugging Face model README
Albedo Qwen3.6-35B — Bookend Phase 4 DPO
LoRA DPO fine-tune on King CXXIV for native bookend emission (Phase 4).
Training
| Param | Value |
|---|---|
| Base | local_king/king_cxxiv (King CXXIV) |
| Pairs | data/dpo/bookend_dpo_pairs.jsonl (1000 pairs) |
| Epochs | 1 (16 steps) |
| LR | 5e-6 |
| Beta | 0.1 |
| LoRA targets | in_proj_qkv, out_proj, q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
| Max length | 16384 |
| Final adapter | checkpoints/bookend_phase4_lora/checkpoint-16 |
| Train loss | ~5.06 |
| Rewards accuracy | ~74% |
Bundle layout
- DPO-tuned language weights (merged from checkpoint-16)
- Unchanged vision weights from King CXXIV
- Genesis metadata (validator
metadata_hashcompatible)
Built with offline merge_lora.py + shard repack per docs/FIELD_NOTES_TRAINING_AND_SUBMIT.md.
- Type
- qwen3_5_moe
- Created
- Sep 9, 2026
- Updated
- Sep 9, 2026