v0.7.1

ossf

Sync with llama2.c

• synced with llama2.c of 2023-08-19 https://github.com/karpathy/llama2.c/tree/bd182289c596fa6059eb7b3b7c8ccd04b5c90fc3
• removed charts
• added links to papers
• fuzz test for top-p sampling

Sync from llama2.c

• sync with llama2.c: karpathy/llama2.c@3c3b19b
• added architecture docs

bugfix key cache size in transformer

issue reported by @datastream

optimisations Fuzz tests and enabling by default

v0.5.0

fuzz

paralelism by goroutines

• 1.5x to 2x speedup

added prompt

v0.3.0

fix

🔥 works with high temperature

• bugfix high temperatures. was multiplying wrong matrix in final RMS step in transformer
• test for mat mul
• nits

update default temperature to 0.1

v0.1.1

Update README.md

works with low temperature

• deterministic (zero temperature, full argmax) works
• low temperature works
• reading from binary checkpoint works
• (bug) high temperature is unstable