pseudotensor's picture
Update README.md
0a7239d
|
raw
history blame
261 Bytes
metadata
license: llama2

Same as h2oai/h2ogpt-16k-codellama-34b-instruct but with config.json modified to be 32k for embeddings, which still functions fine as 16k model and allows stretching into 32k in vLLM that otherwise cannot modify maximum sequence length.