Update README.md
Browse files
README.md
CHANGED
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
base_model:
|
4 |
+
- mistralai/Mistral-Small-24B-Base-2501
|
5 |
+
---
|
6 |
+
|
7 |
+
# Arcee Blitz Caller Beta
|
8 |
+
|
9 |
+
This repository contains a beta version of the Arcee Blitz Caller, a powerful tool for automated function calling and tool selection. Below you'll find instructions on how to launch the service using vllm as well as detailed performance metrics.
|
10 |
+
|
11 |
+
## Overview
|
12 |
+
|
13 |
+
Arcee Blitz Caller is built on top of the Mistral-Small-24B-Base-2501 model and is designed for efficient and accurate tool calling operations.
|
14 |
+
|
15 |
+
## Base Model
|
16 |
+
|
17 |
+
- [mistralai/Mistral-Small-24B-Base-2501](https://huggingface.co/mistralai/Mistral-Small-24B-Base-2501)
|
18 |
+
|
19 |
+
## Quick Start
|
20 |
+
|
21 |
+
To launch the service using vllm, run the following command:
|
22 |
+
|
23 |
+
```bash
|
24 |
+
vllm serve arcee-ai/Arcee-Blitz-Caller \
|
25 |
+
--enable-auto-tool-choice \
|
26 |
+
--tool-call-parser hermes
|
27 |
+
```
|
28 |
+
|
29 |
+
## Performance Metrics
|
30 |
+
|
31 |
+
### Non-Live Performance
|
32 |
+
| Metric | Score |
|
33 |
+
|--------|--------|
|
34 |
+
| AST Accuracy | 85.15% |
|
35 |
+
| Simple AST | 70.08% |
|
36 |
+
| Multiple AST | 93.50% |
|
37 |
+
| Parallel AST | 89.50% |
|
38 |
+
| Parallel Multiple AST | 87.50% |
|
39 |
+
|
40 |
+
### Live Performance
|
41 |
+
| Metric | Score |
|
42 |
+
|--------|--------|
|
43 |
+
| Accuracy | 74.19% |
|
44 |
+
| Simple AST | 71.71% |
|
45 |
+
| Multiple AST | 75.02% |
|
46 |
+
| Parallel AST | 43.75% |
|
47 |
+
| Parallel Multiple AST | 66.67% |
|
48 |
+
|
49 |
+
### Multi-Turn Performance
|
50 |
+
| Metric | Score |
|
51 |
+
|--------|--------|
|
52 |
+
| Accuracy | 0.25% |
|
53 |
+
| Base | 0.50% |
|
54 |
+
| Miss Function | 0.00% |
|
55 |
+
| Miss Parameter | 0.00% |
|
56 |
+
| Long Context | 0.50% |
|
57 |
+
| Relevance Detection | 61.11% |
|
58 |
+
| Irrelevance Detection | 77.68% |
|
59 |
+
|
60 |
+
> **Note:** The model demonstrates strong performance in single-turn interactions but currently shows limitations in multi-turn scenarios. We are actively developing new training data to enhance multi-turn capabilities.
|
61 |
+
|
62 |
+
## Current Status
|
63 |
+
|
64 |
+
This is a beta release and is under active development. While the model shows promising results in many areas, users should be aware of its current limitations, particularly in multi-turn interactions.
|
65 |
+
|
66 |
+
## License
|
67 |
+
|
68 |
+
This project is licensed under the Apache-2.0 License - see the [LICENSE](LICENSE) file for details.
|