Unverified50% confidenceReleaseExact time
The MLX backend of Ollama v0.40.10 supports speculative decoding using a DFlash draft model for the Gemma 4 target model.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/23/2026
First Seen
Valid until: 1/21/2027
Sources
Related Entities
Related Claims
UnverifiedGemma 4 E2B was LoRA-fine-tuned and exported as Q4_K_M GGUF (3.4 GB), run on CPU by llama.cpp's llama-server, to draft two candidate replies and filter memories.66% similarUnverifiedThe models participating in this comparative evaluation include Fable 5, DXU E4GA, DXU E4 Slash N731 official release, Qwen 3 MiniMax 0720, and DeepSeek V4 Pro.65% similarUnverifiedOllama v0.40.4 contains three changes directly related to MLX: state compaction after speculative rollback and an MLX / MLX-C version upgrade, both submitted by dhiltgen.63% similarUnverifiedBecause it uses open-weight models, when the author's local GPU driver failed, they pointed the calls to the hosted version of the same Gemma 4 family (26B) by changing environment variables, continuing without rewriting any code.63% similarUnverifiedOllama has native support for MLX with deep optimization, making it the optimal inference framework choice for M-series chip Macs.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1090549API
curl https://kongchang.com/api/v1/knowledge/claims/1090549MCP
get_claim(id=1090549)