Gemma-4-E4B-DECKARD-HERETIC-NVFP4

NVFP4-quantized EAGLE-style speculative-decoding drafter for the standard (non-uncensored) Gemma 4 31B DECKARD HERETIC. Pair with the matching target model for accelerated single-stream decode on Blackwell-class GPUs (DGX Spark / RTX PRO 6000 / RTX 5090 / B100 / B200).

For the abliterated/uncensored variant of this drafter, see AEON-7/Gemma-4-E4B-DECKARD-HERETIC-Uncensored-NVFP4. For the target it accelerates, see the gemma-4-31B-it-speculator.eagle3-NVFP4 collection on this profile.

Files

• model.safetensors — NVFP4-quantized weights
• hf_quant_config.json — modelopt quant config
• chat_template.jinja — Gemma 4 chat template
• config.json / generation_config.json / tokenizer.* / processor_config.json

Quick start (vLLM, as drafter)

vllm serve <target-model-id> \
  --speculative-config '{"method":"eagle3","model":"AEON-7/Gemma-4-E4B-DECKARD-HERETIC-NVFP4","num_speculative_tokens":3}' \
  --trust-remote-code

License

Inherits the Gemma Terms of Use. Use of this model is subject to those terms.

---

☕ Support the work

AEON-7 models, drafters and tools are built and trained independently, on my own hardware. If they're useful to you, please consider supporting development:

Become a member on Patreon → patreon.com/cw/AeonForge7/membership

Milestones unlock bigger work: reaching 500 paid supporters will fund fine-tuning larger models and bigger project releases. Supporters also get early access to new releases, such as the AEON DFlash2 drafter.

If this release has been useful, tips are deeply appreciated — they go directly toward more compute, more models, and more open releases.

Ethereum L2s (Base, Arbitrum, Optimism, Polygon, etc.) and EVM-compatible tokens can be sent to the same Ethereum address.