From 81f031410d7f75fc9aee844051efa2abf5c2a873 Mon Sep 17 00:00:00 2001 From: avtc Date: Sat, 15 Mar 2025 03:19:33 +0200 Subject: [PATCH 1/2] Update refact_self_hosting.yaml BYOK self-hosting setup. - fix embedding model initialization - use LM Studio for chat model --- .../engine/bring_your_own_key/refact_self_hosting.yaml | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml b/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml index 28681d4cc..bb4790fd1 100644 --- a/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml +++ b/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml @@ -4,8 +4,15 @@ cloud_name: Refact local self-hosting server #chat_endpoint: "http://localhost:8008/v1/chat/completions" #chat_model: "qwen2.5/coder/1.5b/instruct" +# Example of self-hosting via LM Studio +chat_endpoint: "http://localhost:1234/v1/chat/completions" +chat_model: gpt-4o +# Be sure to load any model you prefer in LM Studio first, it will not check the name of a model set there, +# but using "known" for Refact model name like "gpt-4o" will allow use not only "Quick", but also "Explore" and "Agent" modes. + embedding_endpoint: "http://localhost:8008/v1/embeddings" -embedding_model: "thenlper/gte-base" +embedding_model: "thenlper/gte-base" # or "thenlper/gte-base/cpu" +embedding_size: 768 # mandatory to be set completion_endpoint: "http://localhost:8008/v1/completions" completion_model: "Refact/1.6B" From 6de0886d0a17e3446244c94174f8291e6e9e90da Mon Sep 17 00:00:00 2001 From: avtc Date: Sat, 15 Mar 2025 19:17:25 +0200 Subject: [PATCH 2/2] Update refact_self_hosting.yaml --- .../engine/bring_your_own_key/refact_self_hosting.yaml | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml b/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml index bb4790fd1..db38e82d4 100644 --- a/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml +++ b/refact-agent/engine/bring_your_own_key/refact_self_hosting.yaml @@ -6,9 +6,10 @@ cloud_name: Refact local self-hosting server # Example of self-hosting via LM Studio chat_endpoint: "http://localhost:1234/v1/chat/completions" -chat_model: gpt-4o +chat_model: o3-mini # Be sure to load any model you prefer in LM Studio first, it will not check the name of a model set there, -# but using "known" for Refact model name like "gpt-4o" will allow use not only "Quick", but also "Explore" and "Agent" modes. +# but using "known" for Refact model name like "o3-mini" will allow use not only "Quick", but also "Explore" and "Agent" modes. +# for gpt-4o deep_think tool does not work, for o3-mini works embedding_endpoint: "http://localhost:8008/v1/embeddings" embedding_model: "thenlper/gte-base" # or "thenlper/gte-base/cpu"