VisualInference commited on
Commit
7d6821f
·
verified ·
1 Parent(s): e0b70af

Card: fetch the tool-call patch by URL; note the end-to-end check

Browse files
Files changed (1) hide show
  1. README.md +7 -1
README.md CHANGED
@@ -114,10 +114,16 @@ Inkling 2: the model writes a correct call, and the server returns neither `tool
114
  ```sh
115
  git clone https://github.com/Neywa-Labs/mlx-vlm && cd mlx-vlm
116
  git checkout 111ac668078e017c9db1eb3936f4d7a6e0a88f89
117
- git apply /path/to/extras/mlx-vlm-inkling2-tool-calls.patch
 
 
118
  pip install -e .
119
  ```
120
 
 
 
 
 
121
  Everything in the tool and agent rows above was measured with that patch applied. Without a server you can parse
122
  tool calls yourself: generate with `skip_special_tokens=False` and call
123
  `processor.tokenizer.parse_response(text, prefix="")`, which uses the response template shipped in
 
114
  ```sh
115
  git clone https://github.com/Neywa-Labs/mlx-vlm && cd mlx-vlm
116
  git checkout 111ac668078e017c9db1eb3936f4d7a6e0a88f89
117
+ curl -L -o inkling2-tools.patch \
118
+ https://huggingface.co/VisualInference/Inkling-2-Mini-MLX-3bit/resolve/main/extras/mlx-vlm-inkling2-tool-calls.patch
119
+ git apply inkling2-tools.patch
120
  pip install -e .
121
  ```
122
 
123
+ (`load()` downloads only model files, so the patch is not in the local snapshot; fetch it by URL as above.) These
124
+ steps were run against this repository as published: a clean checkout, the patch from the Hub, and a request with
125
+ `tools` returned two parallel `tool_calls` with `finish_reason: "tool_calls"`.
126
+
127
  Everything in the tool and agent rows above was measured with that patch applied. Without a server you can parse
128
  tool calls yourself: generate with `skip_special_tokens=False` and call
129
  `processor.tokenizer.parse_response(text, prefix="")`, which uses the response template shipped in