llm, continue tuning the model context and tokens to prevent swapping.
This commit is contained in:
@@ -591,8 +591,8 @@
|
|||||||
[providers.models.ollama.{{ llm.provider }}]
|
[providers.models.ollama.{{ llm.provider }}]
|
||||||
model = "{{ llm.model }}"
|
model = "{{ llm.model }}"
|
||||||
timeout_secs = 3600 # 1 hour, default is 120 (2 minutes), issue happens between 5 and 6 minutes though.
|
timeout_secs = 3600 # 1 hour, default is 120 (2 minutes), issue happens between 5 and 6 minutes though.
|
||||||
context_window = {{ [llm_context, llm.context] | min | int }}
|
context_window = {{ [llm_context, llm.context, 24000] | min | int }}
|
||||||
max_tokens = {{ [llm_tokens, (llm.context / 4), 16000] | min | int }}
|
max_tokens = {{ [llm_tokens, (llm.context / 4), 4000] | min | int }}
|
||||||
|
|
||||||
{% endif %}
|
{% endif %}
|
||||||
{% endif %}
|
{% endif %}
|
||||||
@@ -625,14 +625,14 @@
|
|||||||
provider: ollama
|
provider: ollama
|
||||||
model: {{ llm.model }}
|
model: {{ llm.model }}
|
||||||
defaultCompletionOptions:
|
defaultCompletionOptions:
|
||||||
contextLength: {{ [llm_context, llm.context] | min | int }}
|
contextLength: {{ [llm_context, llm.context, 32000] | min | int }}
|
||||||
maxTokens: {{ [llm_tokens, (llm.context / 4), 16000] | min | int }}
|
maxTokens: {{ [llm_tokens, (llm.context / 4), 8000] | min | int }}
|
||||||
|
|
||||||
- name: {{ llm.provider }}_autocomplete
|
- name: {{ llm.provider }}_autocomplete
|
||||||
provider: ollama
|
provider: ollama
|
||||||
model: {{ llm.model }}
|
model: {{ llm.model }}
|
||||||
defaultCompletionOptions:
|
defaultCompletionOptions:
|
||||||
contextLength: {{ [llm_context, llm.context] | min | int }}
|
contextLength: {{ [llm_context, llm.context, 16000] | min | int }}
|
||||||
maxTokens: {{ [llm_tokens, (llm.context / 4), 2000] | min | int }}
|
maxTokens: {{ [llm_tokens, (llm.context / 4), 2000] | min | int }}
|
||||||
roles:
|
roles:
|
||||||
- autocomplete
|
- autocomplete
|
||||||
|
|||||||
Reference in New Issue
Block a user