Nanbeige-4.3-33B-EXL3
ExLlamaV3 (EXL3) quantizations of Nanbeige/Nanbeige4.3-33B (33B dense causal LM, 262k context window, native <think> reasoning).
Notes
- Patched
"add_bos_token": falseintokenizer_config.jsonto fix upstream double<|im_start|>BOS token emission when applying chat templates. - Quantized using ExLlamaV3 v1.6.4 with high-quality promotion (
-hq) and codebookmul1. - Uncensored variant: Luni/Nanbeige-4.3-33B-Heretic-EXL3.
Released Branches
| Branch | Avg BPW | Head (-hb) |
Embed (-eb) |
Size | Target VRAM | Notes |
|---|---|---|---|---|---|---|
5.00bpw |
5.00 |
6 |
16 (FP16) |
21.0 GB |
32 GB |
High precision (64k+ context) |
4.50bpw |
4.50 |
6 |
16 (FP16) |
19.3 GB |
24 GB |
KLD-optimal Pareto merge (measure_model + optimize_model) |
4.00bpw |
4.00 |
6 |
16 (FP16) |
17.6 GB |
24 GB / 32 GB |
Standard 4-bit (32k-64k+ context) |
Download & Usage
# Download via huggingface-cli
hf download Luni/Nanbeige-4.3-33B-EXL3 --revision 4.00bpw --local-dir ./Nanbeige-4.3-33B-EXL3-4.00bpw
Model Lineage & Related Repositories
- Base Model: Nanbeige/Nanbeige4.3-33B by Nanbeige Team (Apache-2.0)
- Heretic BF16: Luni/Nanbeige-4.3-33B-Heretic
- Heretic EXL3 Master Suite: Luni/Nanbeige-4.3-33B-Heretic-EXL3
- Heretic EXL3-SC Dedicated Repo: Luni/Nanbeige-4.3-33B-Heretic-EXL3-SC
- Speculative Drafter: Luni/Nanbeige-4.3-33B-DFlash2
- Heretic GGUF & GGUF-IQ Suite: Luni/Nanbeige-4.3-33B-Heretic-GGUF
- Heretic GSQ-RCO Non-Uniform GGUF Suite: Luni/Nanbeige-4.3-33B-Heretic-GSQ-RCO-GGUF