Create README.md

README.md

@@ -0,0 +1,326 @@
+---
+license: other
+license_name: katanemo-research
+license_link: >-
+  https://huggingface.co/katanemolabs/Arch-Function-Calling-1.5B/blob/main/LICENSE
+base_model:
+- Qwen/Qwen2.5-1.5B-Instruct
+---
+
+# Katanemo/Arch-Function-Calling-1.5B
+
+## Overview
+The Katanemo Arch-Function-Calling collection of large language models (LLMs) is a collection state-of-the-art (SOTA) LLMs specifically designed for **function calling** tasks. The models are designed to understand complex function signatures, identify required parameters, and produce accurate function call outputs based on natural language prompts. Achieving performance on par with GPT-4, these models set a new benchmark in the domain of function-oriented tasks, making them suitable for scenarios where automated API interaction and function execution is crucial.
+
+In summary, the Katanemo Arch-Function-Calling collection demonstrates:
+- **State-of-the-art performance** in function calling
+- **Accurate parameter identification and suggestion**, even in ambiguous or incomplete inputs
+- **High generalization** across multiple function calling use cases, from API interactions to automated backend tasks.
+- Optimized **low-latency, high-throughput** performance, making it suitable for real-time, production environments.
+
+
+## Key Features
+<table>
+  <tr style="text-align: left; vertical-align: middle; font-weight: bold;">
+    <td>Functionality</td>
+    <td>Definition</td>
+  </tr>
+  <tr style="text-left: left; vertical-align: middle;">
+    <td>Single Function Calling</td>
+    <td>Call only one function per user query </td>
+  </tr>
+  <tr style="text-left: left; vertical-align: middle;">
+    <td>Parallel Function Calling</td>
+    <td>Call the same function multiple times but with different set of parameter values</td>
+  </tr>
+  <tr style="text-left: left; vertical-align: middle;">
+    <td>Multiple Function Calling</td>
+    <td>Call different functions per user query</td>
+  </tr>
+  <tr style="text-left: left; vertical-align: middle;">
+    <td>Parallel & Multiple</td>
+    <td>Perform both parallel and multiple function calling</td>
+  </tr>
+</table>
+
+
+## Training Details
+Katanemo Arch-Function-Calling collection is built on top of the [Qwen 2.5](https://huggingface.co/collections/Qwen/qwen25-66e81a666513e518adb90d9e). A blog with technical details leading to our models will be published soon.
+
+
+## Performance Benchmarks
+We evaluate Katanemo Arch-Function-Calling series on the [Berkeley Function-Calling Leaderboard (BFCL)](https://gorilla.cs.berkeley.edu/leaderboard.html#leaderboard). For each model family, we select the one with the highest rank. The results are shwon below:
+
+<table>
+  <tr style="text-align: center; vertical-align: middle; font-weight: bold;">
+    <td rowspan=2>Rank</td>
+    <td rowspan=2>Model</td>
+    <td rowspan=2>Overall</td>
+    <td colspan=3>Single Turn</td>
+    <td rowspan=1>Multi Turn</td>
+    <td colspan=2>Hallucination</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle; font-weight: bold;">
+    <td>Non-live (AST)</td>
+    <td>Non-live (Exec)</td>
+    <td>Live (AST)</td>
+    <td>Overall</td>
+    <td>Relevance</td>
+    <td>Irrelevance</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle;">
+    <td>1</td>
+    <td>GPT-4-turbo-2024-04-09</td>
+    <td>59.49%</td>
+    <td>82.65%</td>
+    <td>83.80%</td>
+    <td>73.39%</td>
+    <td>21.62%</td>
+    <td>70.73%</td>
+    <td>79.79%</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle;">
+    <td>3</td>
+    <td>xLAM-8x22b-r</td>
+    <td>59.13%</td>
+    <td>89.75%</td>
+    <td>89.32%</td>
+    <td>72.81%</td>
+    <td>15.62%</td>
+    <td>97.56%</td>
+    <td>75.23%</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle; font-weight: bold;">
+    <td> </td>
+    <td>Arch-Function-Calling-7B</td>
+    <td>57.48%</td>
+    <td>87.50%</td>
+    <td>86.80%</td>
+    <td>72.19%</td>
+    <td>13.75%</td>
+    <td>82.93%</td>
+    <td>79.54%</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle; font-weight: bold;">
+    <td> </td>
+    <td>Arch-Function-Calling-3B</td>
+    <td>56.23%</td>
+    <td>85.10%</td>
+    <td>89.16%</td>
+    <td>70.72%</td>
+    <td>12.28%</td>
+    <td>90.24%</td>
+    <td>73.98%</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle;">
+    <td>7</td>
+    <td>mistral-large-2407</td>
+    <td>55.82%</td>
+    <td>84.12%</td>
+    <td>83.09%</td>
+    <td>67.17%</td>
+    <td>20.50%</td>
+    <td>78.05%</td>
+    <td>48.93%</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle;">
+    <td>9</td>
+    <td>Claude-3.5-Sonnet-20240620</td>
+    <td>54.83%</td>
+    <td>70.35%</td>
+    <td>66.34%</td>
+    <td>71.39%</td>
+    <td>23.5%</td>
+    <td>63.41%</td>
+    <td>75.91%</td>
+  </tr>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle; font-weight: bold;">
+    <td> </td>
+    <td>Arch-Function-Calling-1.5B</td>
+    <td>53.61%</td>
+    <td>82.60%</td>
+    <td>87.36%</td>
+    <td>68.19%</td>
+    <td>8.62%</td>
+    <td>87.80%</td>
+    <td>75.90%</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle;">
+    <td>11</td>
+    <td>o1-mini-2024-09-12</td>
+    <td>53.43%</td>
+    <td>75.48%</td>
+    <td>76.86%</td>
+    <td>71.17%</td>
+    <td>11.00%</td>
+    <td>46.34%</td>
+    <td>88.07%</td>
+  </tr>
+  <tr style="text-align: center; vertical-align: middle;">
+    <td>12</td>
+    <td>Gemini-1.5-Flash-Preview-0514</td>
+    <td>53.01%</td>
+    <td>77.10%</td>
+    <td>71.23%</td>
+    <td>71.17%</td>
+    <td>13.12%</td>
+    <td>60.98%</td>
+    <td>76.15%</td>
+  </tr>
+</table>
+
+
+# Requirements
+The code of Arch-Function-Calling-1.5B has been in the latest Hugging face transformers and we advise you to install install the `transformers` library:
+```bash
+pip install transformers>=4.37.0
+```
+
+
+# How to use
+We use the following example to illustrate how to use our model to perform function calling tasks. Please note that, our model works best with our provided prompt format. It allows us to extract JSON output that is similar to the [function-calling mode of ChatGPT](https://platform.openai.com/docs/guides/function-calling).
+
+
+### Single Turn Example
+````python
+import json
+from typing import Any, Dict, List
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model_name = "katanemolabs/Arch-Function-Calling-1.5B"
+model = AutoModelForCausalLM.from_pretrained(
+    model_name, device_map="auto", torch_dtype="auto", trust_remote_code=True
+)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+# Please use our provided prompt for best performance
+TASK_PROMPT = """
+You are a helpful assistant.
+""".strip()
+
+TOOL_PROMPT = """
+# Tools
+
+You may call one or more functions to assist with the user query.
+
+You are provided with function signatures within <tools></tools> XML tags:
+<tools>
+{tool_text}
+</tools>
+""".strip()
+
+FORMAT_PROMPT = """
+For each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:
+<tool_call>
+{"name": <function-name>, "arguments": <args-json-object>}
+</tool_call>
+""".strip()
+
+# Define available tools
+get_weather_api = {
+    "type": "function",
+    "function": {
+        "name": "get_weather",
+        "description": "Get the current weather for a location",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "location": {
+                    "type": "str",
+                    "description": "The city and state, e.g. San Francisco, New York",
+                },
+                "unit": {
+                    "type": "str",
+                    "enum": ["celsius", "fahrenheit"],
+                    "description": "The unit of temperature to return",
+                },
+            },
+            "required": ["location"],
+        },
+    },
+}
+
+openai_format_tools = [get_weather_api]
+
+
+def convert_tools(tools: List[Dict[str, Any]]):
+    return "\n".join([json.dumps(tool) for tool in tools])
+
+# Helper function to create the system prompt for our model
+def format_prompt(tools: List[Dict[str, Any]]):
+    tool_text = convert_tools(tools)
+
+    return (
+        TASK_PROMPT
+        + "\n\n"
+        + TOOL_PROMPT.format(tool_text=tool_text)
+        + "\n\n"
+        + FORMAT_PROMPT
+        + "\n"
+    )
+
+
+system_prompt = format_prompt(openai_format_tools)
+
+messages = [
+    {"role": "system", "content": system_prompt},
+    {"role": "user", "content": "What is the weather in Seattle?"},
+]
+
+inputs = tokenizer.apply_chat_template(
+    messages, add_generation_prompt=True, return_tensors="pt"
+).to(model.device)
+
+# tokenizer.eos_token_id is the id of <|EOT|> token
+outputs = model.generate(
+    inputs,
+    max_new_tokens=512,
+    do_sample=False,
+    num_return_sequences=1,
+    eos_token_id=tokenizer.eos_token_id,
+)
+
+response = tokenizer.decode(outputs[0][len(inputs[0]) :], skip_special_tokens=True)
+print(response)
+````
+
+Then you should be able to see the following output string in JSON format:
+````python
+<tool_call>
+{"name": "get_weather", "arguments": {"location": "Seattle"}}
+</tool_call>
+````
+
+### Multi Turn Example
+Upon getting results from functions, you can add it to the `messages` list as a `user` message and pass it to the model to get responses for users.
+
+````python
+# Suppose we receive the following result from the function:
+get_weather_api_result = {'name': 'get_weather', 'results': {'temperature': '62°', 'unit': 'fahrenheit'}}
+execution_results = [get_weather_api_result]
+
+def add_execution_results(messages: List[Dict[str, Any]], execution_results: List[Dict[str, Any]]):
+    content = "\n".join([f"<tool_response>\n{json.dumps(result)}</tool_response>" for result in execution_results])
+    messages.append({"role": "user", "content": content})
+    return messages
+
+messages = add_execution_results(messages, execution_results)
+
+inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
+
+outputs = model.generate(
+    inputs,
+    max_new_tokens=512,
+    do_sample=False,
+    num_return_sequences=1,
+    eos_token_id=tokenizer.eos_token_id,
+)
+
+response = tokenizer.decode(outputs[0][len(inputs[0]) :], skip_special_tokens=True)
+print(response)
+````
+
+
+# License
+Katanemo Arch-Function-Calling collection is distributed under the [Katanemo license](https://huggingface.co/katanemolabs/Arch-Function-Calling-1.5B/blob/main/LICENSE).