What if you used the new unreal MCP so a lot of the transforms and blocking out even is doing via voice and you can just tweak it
LLMS
-
Claude Code burning 2x tokens of Codex is spiciest line
By
–
Claude Code burning 2x the tokens of Codex is the spiciest line in here haha.
-
No speed difference, but Ollama offers nice quantized models and easy switching
By
–
It's been a few months but I didn't see a noticeable speed difference tbh. Otherwise, Ollama has usually a nice set of quantized models and easy to switch.
-
Bad news: GPT-5.6 global rollout uncertain
By
–
Ive got bad news for you: it’s not certain GPT-5.6 will be rolled out globally
-
Decode 80 tok/sec at 1k, prefill 3000 tok/sec at 50k contexts
By
–
Yeah, this was at 50k contexts. Decode is about 80 tok/sec at 1k contexts. Prefill is up to 3000 tok/sec at
-
DeepSeek uses open-perfectblend dataset to train DSpark drafter
By
–
Fun surprise: DeepSeek used my open-perfectblend dataset to train their new DSpark drafter Time to promote it again! It's an open-source reproduction of "The Perfect Blend" paper. If you ever need >1M diverse prompts in math, chat, and code, it does the job.
-
hf-claude lets you use over 100 open models in Claude Code
By
–
hf-claude lets you use over 100 open models in claude code including glm 5.2, minimax-m3, deepseek v4 pro
-
LMCache offloads KV cache to CPU, disk, or S3 instead of evicting
By
–
Compression is one way to fight the KV cache wall. The other is to not throw the KV away. LMCache reuses and offloads it to CPU, disk, or S3 instead of evicting. Already plugs into vLLM, SGLang, and NVIDIA Dynamo. Worth a star if you serve LLMs.
-
Change the base model of an LLM in one line
By
–
With any serious LLM library, it takes one line to change the core model.
