name: Codestreal 22B
model: codestral:22B
version: 1

files:
	- llama_model_path: model.gguf

# Results Preferences
top_p: 0.95
temperature: 0.7
frequency_penalty: 0
presence_penalty: 0
max_tokens: 32000 # Infer from base config.json -> max_position_embeddings
stream: true # true | false

# Engine / Model Settings
ngl: 32 # Infer from base config.json -> num_attention_heads
ctx_len: 32000 # Infer from base config.json -> max_position_embeddings
engine: cortex.llamacpp
prompt_template: "{system_message} [INST] {prompt} [/INST]"
# Prompt template: Can only be retrieved from instruct model
# - https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct/blob/main/tokenizer_config.json#L2053
# - Requires jinja format parser