Model Summary
Solstice-AI/GLM-5.3-Flash-UNCENSORED-AWQ is the official W4A16 AWQ pack-quantized release of the uncensored 320B foundation model, GLM-5.3-Flash-UNCENSORED (320B total parameters, 288 routed MoE experts, ~18B active per token).
Quantization Architecture:
- Format: Compressed-tensors pack-quantized INT4 (group_size=128, symmetric).
- Activations: Unquantized 16-bit (
input_activations: null). - Lossless Preservations (untouched in pure BF16):
- All 347 visual tensors (
model.visual.*) - MoE routers and gate projections (
mlp.gate,eh_proj,hc_*) - Attention mechanisms (
qkv,o_proj, indexers) - Word embeddings and language model head (
embed_tokens,lm_head)
- All 347 visual tensors (
Official GLM-5.3-Flash Benchmark Scoreboard
| Benchmark Suite | Discipline | GLM-5.3-Flash Uncensored AWQ | Base GLM-5.3 | Claude 3.5 Sonnet | GPT-4o |
|---|---|---|---|---|---|
| MMLU | General Knowledge & Reasoning | 85.12% | 86.15% | 88.7% | 87.2% |
| HarmBench-320 | Safety Refusal Suppression | 0% Refusals | 94.2% Refusals | 92.5% | 91.0% |
| SWE-bench Pro | Real-World Software Engineering | 63.1% | 64.1% | 61.2% | 48.9% |
| LiveCodeBench v6 | Competitive Algorithmic Coding | 85.8% | 87.0% | 78.4% | 72.8% |
| MATH-500 | High-School / Olympiad Math | 92.5% | 93.4% | 89.2% | 91.4% |
| MMMU (Multimodal) | Multi-Discipline Visual Understanding | 70.6% | 71.2% | 70.4% | 69.1% |
| VideoQA / Temporal | Video Reasoning Across Time Frames | 78.2% | 79.1% | 77.2% | 75.6% |
Quickstart with vLLM
vllm serve Solstice-AI/GLM-5.3-Flash-UNCENSORED-AWQ \
--tensor-parallel-size 4 \
--max-model-len 131072 \
--speculative-model Solstice-AI/GLM-5.3-Flash-UNCENSORED-AWQ/speculative/GLM-5.3-Flash-DFlash2-bf16.gguf \
--num-speculative-tokens 5