txgsync commited on
Commit
05c2104
·
verified ·
1 Parent(s): f26e0a2

Document Maple sampler and context settings

Browse files
Files changed (1) hide show
  1. README.md +16 -0
README.md CHANGED
@@ -26,6 +26,22 @@ This repository contains the full-precision BF16 MLX conversion of [`deepgrove/m
26
 
27
  This is an MLX conversion for local inference on Apple Silicon. Please follow the base model's MIT license and usage terms.
28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
29
  ---
30
 
31
  ## Base model description
 
26
 
27
  This is an MLX conversion for local inference on Apple Silicon. Please follow the base model's MIT license and usage terms.
28
 
29
+ ## Recommended generation settings
30
+
31
+ Maple is a reasoning-heavy model and may spend a substantial part of its response budget thinking. For the OpenAI-compatible API or oMLX UI, start with:
32
+
33
+ ```text
34
+ temperature: 1.0
35
+ top_p: 0.95
36
+ top_k: 40
37
+ min_p: 0.05
38
+ repetition_penalty: 1.0
39
+ max_tokens: 8192 or higher
40
+ max context: 131072 tokens (native model limit)
41
+ ```
42
+
43
+ These sampler values match DeepGrove's Maple `llama.cpp` setup. The model declares a native 131,072-token context window and does not require RoPE/YARN scaling for that window. Actual usable context may be lower on systems constrained by KV-cache memory; do not assume that extending beyond 131,072 tokens is supported.
44
+
45
  ---
46
 
47
  ## Base model description