From d65c10cee7120a3706502c6990b287273cf31240 Mon Sep 17 00:00:00 2001 From: Ettore Di Giacinto Date: Tue, 24 Dec 2024 11:04:19 +0100 Subject: [PATCH] chore(model gallery): add tqwendo-36b (#4489) Signed-off-by: Ettore Di Giacinto --- gallery/index.yaml | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/gallery/index.yaml b/gallery/index.yaml index 7c8b1ca30cf6..8191fd27758a 100644 --- a/gallery/index.yaml +++ b/gallery/index.yaml @@ -2313,6 +2313,23 @@ - filename: QwQ-LCoT-7B-Instruct-Q4_K_M.gguf sha256: 1df2e4ff0093a9632687b73969153442776b0ffc1c3c68e7f559472f9cea1945 uri: huggingface://bartowski/QwQ-LCoT-7B-Instruct-GGUF/QwQ-LCoT-7B-Instruct-Q4_K_M.gguf +- !!merge <<: *qwen25 + name: "tqwendo-36b" + icon: "https://cdn-uploads.huggingface.co/production/uploads/6379683a81c1783a4a2ddba8/DI7Yw8Fs8eukluzKTHjEH.png" + urls: + - https://huggingface.co/nisten/tqwendo-36b + - https://huggingface.co/bartowski/tqwendo-36b-GGUF + description: | + There is a draft model to go with this one for speculative decoding and chain of thought reasoning: https://huggingface.co/nisten/qwen2.5-coder-7b-abliterated-128k-AWQ + + Using the above 4bit 7b in conjuction with the 36b is meant to setup a chain-of-thought reasoner, evaluator similar to what O1-O3 is probably doing. This way the 7b 4bit only uses up an extra 4-6Gb on the gpu, but greatly both speeds up speculative decoding AND also chain-of-throught evals. + overrides: + parameters: + model: tqwendo-36b-Q4_K_M.gguf + files: + - filename: tqwendo-36b-Q4_K_M.gguf + sha256: 890ff05fb717c67848d5c02ad62b2c26fdcdd20f7cc94ade8095869784c0cc82 + uri: huggingface://bartowski/tqwendo-36b-GGUF/tqwendo-36b-Q4_K_M.gguf - &smollm ## SmolLM url: "github:mudler/LocalAI/gallery/chatml.yaml@master"