Merge 59487556ce into bc1fca39f3

nanochat/gpt.py

@@ -8,7 +8,7 @@ Notable features:
 - norm after token embedding
 - no learnable params in rmsnorm
 - no bias in linear layers
-- Multi-Query Attention (MQA) support for more efficient inference
+- Group-Query Attention (GQA) support for more efficient inference
 """
 
 import math
@@ -29,7 +29,7 @@ class GPTConfig:
     vocab_size: int = 50304
     n_layer: int = 12
     n_head: int = 6 # number of query heads
-    n_kv_head: int = 6 # number of key/value heads (MQA)
+    n_kv_head: int = 6 # number of key/value heads (GQA)
     n_embd: int = 768
 
 

rustbpe/pyproject.toml

@@ -0,0 +1,18 @@
+[project]
+name = "rustbpe"
+version = "0.1.0"
+description = "Rust BPE tokenizer from nanochat as standalone package"
+readme = "README.md"
+requires-python = ">=3.10"
+license = {text = "MIT"}
+dependencies = []
+
+[build-system]
+requires = ["maturin>=1.0"]
+build-backend = "maturin"
+
+[tool.maturin]
+module-name = "rustbpe"
+bindings = "pyo3"
+python-source = "."
+features = ["pyo3/extension-module"]