mirror of
https://github.com/zenlm/zen-eco.git
synced 2026-10-30 14:58:11 +00:00
105 lines
2.5 KiB
YAML
105 lines
2.5 KiB
YAML
# Zen-Eco Agent Tool Calling Dataset Configuration
|
|
# Dataset: Salesforce/xlam-function-calling-60k
|
|
# Model: Qwen3-4B fine-tuned for function calling
|
|
|
|
dataset:
|
|
name: "xlam-function-calling-60k"
|
|
provider: "Salesforce"
|
|
hf_dataset_id: "Salesforce/xlam-function-calling-60k"
|
|
url: "https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k"
|
|
size: 60000
|
|
license: "Apache 2.0"
|
|
description: |
|
|
High-quality function calling dataset with 60k examples
|
|
Used to fine-tune Qwen3-4B for tool calling capabilities
|
|
Covers diverse function calling scenarios and tool usage patterns
|
|
|
|
features:
|
|
- query: "User query that requires function calling"
|
|
- functions: "Available functions/tools"
|
|
- answers: "Expected function calls and responses"
|
|
- type: "Type of function calling scenario"
|
|
|
|
statistics:
|
|
total_examples: 60000
|
|
unique_functions: 5470
|
|
downloads: 520000
|
|
languages: ["en"]
|
|
task_categories:
|
|
- "function-calling"
|
|
- "tool-use"
|
|
- "api-calling"
|
|
|
|
model:
|
|
base_model: "Qwen/Qwen3-4B-Instruct"
|
|
fine_tuned_model: "Manojb/Qwen3-4b-toolcall-gguf-llamacpp-codex"
|
|
parameters: "4B"
|
|
architecture: "Qwen3"
|
|
training:
|
|
method: "LoRA"
|
|
dataset: "Salesforce/xlam-function-calling-60k"
|
|
epochs: 3
|
|
batch_size: 8
|
|
learning_rate: 2e-5
|
|
lora_rank: 64
|
|
lora_alpha: 128
|
|
lora_dropout: 0.05
|
|
|
|
quantization:
|
|
format: "GGUF"
|
|
method: "Q8_0"
|
|
bits: 8
|
|
file_size: "4.28GB"
|
|
|
|
capabilities:
|
|
max_context: 262144
|
|
supports_tools: true
|
|
supports_parallel_calls: true
|
|
supports_json_mode: true
|
|
|
|
training_config:
|
|
data_preprocessing:
|
|
max_length: 4096
|
|
padding: "max_length"
|
|
truncation: true
|
|
add_special_tokens: true
|
|
|
|
prompt_template: |
|
|
<|im_start|>system
|
|
You are a helpful AI assistant with access to tools.
|
|
{tools_description}
|
|
<|im_end|>
|
|
<|im_start|>user
|
|
{query}
|
|
<|im_end|>
|
|
<|im_start|>assistant
|
|
{response}
|
|
<|im_end|>
|
|
|
|
tool_format:
|
|
type: "json"
|
|
schema:
|
|
name: "string"
|
|
arguments: "object"
|
|
description: "string"
|
|
|
|
evaluation:
|
|
metrics:
|
|
- accuracy: "Tool selection accuracy"
|
|
- f1_score: "Argument extraction F1"
|
|
- execution_rate: "Successful execution rate"
|
|
benchmarks:
|
|
xlam_test: 0.89
|
|
tool_bench: 0.86
|
|
function_call_accuracy: 0.91
|
|
|
|
deployment:
|
|
inference_backend: "llama-cpp-python"
|
|
min_requirements:
|
|
ram: "6GB"
|
|
vram: "4GB (optional)"
|
|
disk: "5GB"
|
|
recommended:
|
|
ram: "8GB"
|
|
vram: "6GB"
|
|
disk: "10GB" |