Views
No views yet
| Metric | llama3.2 (base) | phila-ft-v2 |
|---|---|---|
| Gate accuracy (holdout, 43 scenarios) | 87.9% | 93.0% |
| Gate accuracy (all 101 scenarios) | 94.1% | 95.8% |
| Response quality | 0.951 | 0.965 |
| Composite score | 0.8487 | 0.8638 |
| Avg latency | 515ms | 544ms |
1# download the GGUF
2# then create with the included Modelfile:
3ollama create phila-ft-v2 -f Modelfile
4
5# or manually:
6ollama create phila-ft-v2 -f - <<EOF
7FROM ./phila-ft-v2.Q4_K_M.gguf
8TEMPLATE \"\"\"{{- if .System }}<|start_header_id|>system<|end_header_id|>
9
10{{ .System }}<|eot_id|>{{ end }}{{- range .Messages }}<|start_header_id|>{{ .Role }}<|end_header_id|>
11
12{{ .Content }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|>
13
14\"\"\"
15PARAMETER stop "<|start_header_id|>"
16PARAMETER stop "<|end_header_id|>"
17PARAMETER stop "<|eot_id|>"
18PARAMETER temperature 0.1
19PARAMETER top_p 0.52
20PARAMETER num_predict 64
21EOF