Downloads · 30 days
44
4% of all-time downloads
Danny-Dasilva/Qwen3-8B-antidoom-W4A16-GPTQ-g32-DSpark
Qwen3-8B-antidoom-W4A16-GPTQ-g32-DSpark is a text generation model from Danny-Dasilva. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
An Antidoom (FTPO anti-repetition) version of Qwen/Qwen3-8B, quantized post-training to GPTQ int4 (W4A16, group=32, symmetric) and verified as a DSpark speculative-decoding target (draft: deepseek-ai/dsparkqwen38bbloc…
Downloads · 30 days
44
4% of all-time downloads
All-time downloads
1.1K
Public
Parameters
8.4B
6.4 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors6.4 GB · 100%
How the weights are stored.
I326.9B · 83%
From the Hugging Face model README
An Antidoom (FTPO anti-repetition) version of
Qwen/Qwen3-8B, quantized post-training to
GPTQ int4 (W4A16, group=32, symmetric) and verified as a
DSpark speculative-decoding target
(draft: deepseek-ai/dspark_qwen3_8b_block7).
341 tok/s single-stream on one RTX 5090 (32 GB) with DSpark k=7 + CUDA graphs — at 6.0 GB of weights.
Antidoom FTPO pass: sampled completions at temperature 0.01 from the
LiquidAI/antidoom-mix-v1.0 prompt mix
(filtered to the highest doom-loop-yield sources: code + instruction-following), detected runaway
repetition, extracted 156 FTPO preference pairs (105 after regularisation filtering), trained a LoRA (r=128, all layers + lm_head, max_seq 3072)
with early stop at chosen_win ≥ 0.4, and merged it into the base weights.
Post-training GPTQ (llm-compressor): int4 W4A16 group=32 symmetric, 256 ultrachat calibration samples with chat template. Post-training quantization preserves DSpark draft acceptance — retraining-style (QAT) quantization destroys it (see gemma-4-12B-it-W4A16-GPTQ-g32-DSpark).
DSpark acceptance is preserved through both steps: ~25% (original bf16) → 24.8% (after antidoom) → 26.1% (after GPTQ). The FTPO patch is local and targeted; it does not break the draft head.
| config | tok/s | DSpark accept |
|---|---|---|
| this model + DSpark k=7 + CUDA graphs | 341 (peaks 454) | 26.1% |
| this model, native (no speculation) | 224 | — |
| antidoom merged bf16 + DSpark k=7 | 188 | 24.8% |
from vllm import LLM, SamplingParams
from transformers import AutoTokenizer
MODEL = "Danny-Dasilva/Qwen3-8B-antidoom-W4A16-GPTQ-g32-DSpark"
DRAFT = "deepseek-ai/dspark_qwen3_8b_block7"
llm = LLM(
model=MODEL,
max_model_len=8192,
attention_backend="FLASHINFER",
speculative_config={
"method": "dspark",
"model": DRAFT,
"num_speculative_tokens": 7, # draft block size; keep at 7
"attention_backend": "TRITON_ATTN",
},
)
tok = AutoTokenizer.from_pretrained(MODEL)
prompt = tok.apply_chat_template([{"role": "user", "content": "Hello!"}],
add_generation_prompt=True, tokenize=False)
print(llm.generate([prompt], SamplingParams(temperature=0.0, max_tokens=256))[0].outputs[0].text)
Requires a vLLM build with DSpark support (merged to vLLM main 2026-07-02).
temperature=0.01, prompt sources filtered to code/instruction-following (3-8× higher loop yield
than math/QA), pair budget 156 (generation cut at ~5h; the chosen-win early stop converges regardless), num_epochs=3, lora_r=128, max_seq_length=3072 (fits 32 GB — the FTPO trainer materialises full-vocab logits), with the built-in chosen_win ≥ 0.4 early stop.Qwen/Qwen3-8B (bf16) → antidoom FTPO LoRA merge → GPTQ W4A16 g32