llm-templates/Modelfile-nemotron at main · sammcj/llm-templates · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
#  ollama create nvidia_llama-3_3-nemotron-super-49b-v1:q4_k_l -f modelfiles/Modelfile-nemotron

FROM ../nvidia_Llama-3_3-Nemotron-Super-49B-v1-Q4_K_L.gguf


TEMPLATE """{{- "<|start_header_id|>system<|end_header_id>\n\n" -}}
{{- if and (gt (len .Messages) 0) (eq (index .Messages 0).Role "system") -}}
{{- (index .Messages 0).Content -}}
{{- end -}}
{{- "<|eot_id>" -}}
{{- range $i, $msg := .Messages -}}
{{- if or (gt $i 0) (ne $msg.Role "system") -}}
{{- "<|start_header_id|>" -}}
{{- $msg.Role -}}
{{- "<|end_header_id>\n\n" -}}
{{- $msg.Content -}}
{{- "<|eot_id>" -}}
{{- end -}}
{{- end -}}
{{- "<|start_header_id|>assistant<|end_header_id>\n\n" -}}"""


PARAMETER num_ctx 8192

#  PARAMETER stop "<|start_header_id|>"
#  PARAMETER stop "<|end_header_id|>"
PARAMETER stop "<|eot_id|>"

#  PARAMETER num_batch 1024

## For codegen ##
#  PARAMETER temperature 0.25
#  PARAMETER top_p 0.8

### min_p sampling ##
# min_p works best with a bit of temperature
PARAMETER temperature 0.2
# 1.0 disables top_p, so we can use min_p
PARAMETER top_p 1.0
PARAMETER min_p 0.9
### min_p sampling ##

#  PARAMETER presence_penalty 0.2
#  PARAMETER frequency_penalty 0.2
#  PARAMETER repeat_last_n 50