I've been experimenting with a new quantization approach that selectively elevates the precision of key layers beyond what the default IMatrix configuration provides.
In my testing, standard IMatrix quantization underperforms at lower bit depths, especially with Mixture of Experts (MoE) models. To address this, I'm using the --tensor-type option in llama.cpp to manually "bump" important layers to higher precision. You can see the implementation here:
👉 Layer bumping with llama.cpp
While this does increase model file size, it significantly improves precision for a given quantization level.
I'd love your feedback—have you tried this? How does it perform for you?
Large Action Models (LAMs) are advanced language models designed to enhance decision-making by translating user intentions into executable actions. As the brains of AI agents, LAMs autonomously plan and execute tasks to achieve specific goals, making them invaluable for automating workflows across diverse domains. This model release is for research purposes only.
The new xLAM-2 series, built on our most advanced data synthesis, processing, and training pipelines, marks a significant leap in multi-turn conversation and tool usage. Trained using our novel APIGen-MT framework, which generates high-quality training data through simulated agent-human interactions. Our models achieve state-of-the-art performance on BFCL and τ-bench benchmarks, outperforming frontier models like GPT-4o and Claude 3.5. Notably, even our smaller models demonstrate superior capabilities in multi-turn scenarios while maintaining exceptional consistency across trials.
We've also refined the chat template and vLLM integration, making it easier to build advanced AI agents. Compared to previous xLAM models, xLAM-2 offers superior performance and seamless deployment across applications.
Model Performance Overview
Comparative performance of larger xLAM-2-fc-r models (8B-70B, trained with APIGen-MT data) against state-of-the-art baselines on function-calling (BFCL v3, as of date 04/02/2025) and agentic (τ-bench) capabilities.
xLAM series are significant better at many things including general tasks and function calling.
For the same number of parameters, the model have been fine-tuned across a wide range of agent tasks and scenarios, all while preserving the capabilities of the original model.
*Note: The default context length for Qwen-2.5-based models is 32k, but you can use techniques like YaRN (Yet Another Recursive Network) to achieve maximum 128k context length. Please refer to here for more details.
You can also explore our previous xLAM series here.
The -fc suffix indicates that the models are fine-tuned for function calling tasks, while the -r suffix signifies a research release.
✅ All models are fully compatible with vLLM and Transformers-based inference frameworks.
Usage
Framework versions
Transformers 4.46.1 (or later)
PyTorch 2.5.1+cu124 (or later)
Datasets 3.1.0 (or later)
Tokenizers 0.20.3 (or later)
Basic Usage with Huggingface Chat Template
The new xLAM models are designed to work seamlessly with the Hugging Face Transformers library and utilize natural chat templates for an easy and intuitive conversational experience. Below are examples of how to use these models.
python
1import torch
2from transformers import AutoModelForCausalLM, AutoTokenizer
34tokenizer = AutoTokenizer.from_pretrained("Salesforce/Llama-xLAM-2-3b-fc-r")5model = AutoModelForCausalLM.from_pretrained("Salesforce/Llama-xLAM-2-3b-fc-r", torch_dtype=torch.bfloat16, device_map="auto")67# Example conversation with a tool call8messages =[9{"role":"user","content":"Hi, how are you?"},10{"role":"assistant","content":"Thanks. I am doing well. How can I help you?"},11{"role":"user","content":"What's the weather like in London?"},12]1314tools =[15{16"name":"get_weather",17"description":"Get the current weather for a location",18"parameters":{19"type":"object",20"properties":{21"location":{"type":"string","description":"The city and state, e.g. San Francisco, CA"},22"unit":{"type":"string","enum":["celsius","fahrenheit"],"description":"The unit of temperature to return"}23},24"required":["location"]25}26}27]2829print("====== prompt after applying chat template ======")30print(tokenizer.apply_chat_template(messages, tools=tools, add_generation_prompt=True, tokenize=False))3132inputs = tokenizer.apply_chat_template(messages, tools=tools, add_generation_prompt=True, return_dict=True, return_tensors="pt")33input_ids_len = inputs["input_ids"].shape[-1]# Get the length of the input tokens34inputs ={k: v.to(model.device)for k, v in inputs.items()}35print("====== model response ======")36outputs = model.generate(**inputs, max_new_tokens=256)37generated_tokens = outputs[:, input_ids_len:]# Slice the output to get only the newly generated tokens38print(tokenizer.decode(generated_tokens[0], skip_special_tokens=True))
Using vLLM for Inference
The xLAM models can also be efficiently served using vLLM for high-throughput inference. Please use vllm>=0.6.5 since earlier versions will cause degraded performance for Qwen-based models.
Setup and Serving
Install vLLM with the required version:
pip install "vllm>=0.6.5"
Download the tool parser plugin to your local path:
Note: Ensure that the tool parser plugin file is downloaded and that the path specified in --tool-parser-plugin correctly points to your local copy of the file. The xLAM series models all utilize the same tool call parser, so you only need to download it once for all models.
Testing with OpenAI API
Here's a minimal example to test tool usage with the served endpoint:
python
1import openai
2import json
34# Configure the client to use your local vLLM endpoint5client = openai.OpenAI(6 base_url="http://localhost:8000/v1",# Default vLLM server URL7 api_key="empty"# Can be any string8)910# Define a tool/function11tools =[12{13"type":"function",14"function":{15"name":"get_weather",16"description":"Get the current weather for a location",17"parameters":{18"type":"object",19"properties":{20"location":{21"type":"string",22"description":"The city and state, e.g. San Francisco, CA"23},24"unit":{25"type":"string",26"enum":["celsius","fahrenheit"],27"description":"The unit of temperature to return"28}29},30"required":["location"]31}32}33}34]3536# Create a chat completion37response = client.chat.completions.create(38 model="Salesforce/xLAM-2-1b-fc-r",# Model name doesn't matter, vLLM uses the served model39 messages=[40{"role":"system","content":"You are a helpful assistant that can use tools."},41{"role":"user","content":"What's the weather like in San Francisco?"}42],43 tools=tools,44 tool_choice="auto"45)4647# Print the response48print("Assistant's response:")49print(json.dumps(response.model_dump(), indent=2))
For more advanced configurations and deployment options, please refer to the vLLM documentation.
Benchmark Results
Berkeley Function-Calling Leaderboard (BFCL v3)
BFCL Results
Performance comparison of different models on [BFCL leaderboard](https://gorilla.cs.berkeley.edu/leaderboard.html). The rank is based on the overall accuracy, which is a weighted average of different evaluation categories. "FC" stands for function-calling mode in contrast to using a customized "prompt" to extract the function calls.
τ-bench Benchmark
Tau-bench Results
Success Rate (pass@1) on τ-bench benchmark averaged across at least 5 trials. Our xLAM-2-70b-fc-r model achieves an overall success rate of 56.2% on τ-bench, significantly outperforming the base Llama 3.1 70B Instruct model (38.2%) and other open-source models like DeepSeek v3 (40.6%). Notably, our best model even outperforms proprietary models such as GPT-4o (52.9%) and approaches the performance of more recent models like Claude 3.5 Sonnet (new) (60.1%).
Pass^k curves
Pass^k curves measuring the probability that all 5 independent trials succeed for a given task, averaged across all tasks for τ-retail (left) and τ-airline (right) domains. Higher values indicate better consistency of the models.
Ethical Considerations
This release is for research purposes only in support of an academic paper. Our models, datasets, and code are not specifically designed or evaluated for all downstream purposes. We strongly recommend users evaluate and address potential concerns related to accuracy, safety, and fairness before deploying this model. We encourage users to consider the common limitations of AI, comply with applicable laws, and leverage best practices when selecting use cases, particularly for high-risk scenarios where errors or misuse could significantly impact people's lives, rights, or safety. For further guidance on use cases, refer to our AUP and AI AUP.
If you use our model or dataset in your work, please cite our paper:
bibtex
1@article{prabhakar2025apigen,
2 title={APIGen-MT: Agentic PIpeline for Multi-Turn Data Generation via Simulated Agent-Human Interplay},
3 author={Prabhakar, Akshara and Liu, Zuxin and Zhu, Ming and Zhang, Jianguo and Awalgaonkar, Tulika and Wang, Shiyu and Liu, Zhiwei and Chen, Haolin and Hoang, Thai and others},
4 journal={arXiv preprint arXiv:2504.03601},
5 year={2025}
6}
Additionally, please check our other awesome related works regarding xLAM series and consider citing them as well:
bibtex
1@article{zhang2025actionstudio,
2 title={ActionStudio: A Lightweight Framework for Data and Training of Action Models},
3 author={Zhang, Jianguo and Hoang, Thai and Zhu, Ming and Liu, Zuxin and Wang, Shiyu and Awalgaonkar, Tulika and Prabhakar, Akshara and Chen, Haolin and Yao, Weiran and Liu, Zhiwei and others},
4 journal={arXiv preprint arXiv:2503.22673},
5 year={2025}
6}
bibtex
1@article{zhang2024xlam,
2 title={xLAM: A Family of Large Action Models to Empower AI Agent Systems},
3 author={Zhang, Jianguo and Lan, Tian and Zhu, Ming and Liu, Zuxin and Hoang, Thai and Kokane, Shirley and Yao, Weiran and Tan, Juntao and Prabhakar, Akshara and Chen, Haolin and others},
4 journal={arXiv preprint arXiv:2409.03215},
5 year={2024}
6}
7
bibtex
1@article{liu2024apigen,
2 title={Apigen: Automated pipeline for generating verifiable and diverse function-calling datasets},
3 author={Liu, Zuxin and Hoang, Thai and Zhang, Jianguo and Zhu, Ming and Lan, Tian and Tan, Juntao and Yao, Weiran and Liu, Zhiwei and Feng, Yihao and RN, Rithesh and others},
4 journal={Advances in Neural Information Processing Systems},
5 volume={37},
6 pages={54463--54482},
7 year={2024}
8}
bibtex
1@article{zhang2024agentohana,
2 title={AgentOhana: Design Unified Data and Training Pipeline for Effective Agent Learning},
3 author={Zhang, Jianguo and Lan, Tian and Murthy, Rithesh and Liu, Zhiwei and Yao, Weiran and Tan, Juntao and Hoang, Thai and Yang, Liangwei and Feng, Yihao and Liu, Zuxin and others},
4 journal={arXiv preprint arXiv:2402.15506},
5 year={2024}
6}
🚀 If you find these models useful
Help me test my AI-Powered Quantum Network Monitor Assistant with quantum-ready security checks:
The full Open Source Code for the Quantum Network Monitor Service available at my github repos ( repos with NetworkMonitor in the name) : Source Code Quantum Network Monitor. You will also find the code I use to quantize the models if you want to do it yourself GGUFModelBuilder
💬 How to test:
Choose an AI assistant type:
TurboLLM (GPT-4.1-mini)
HugLLM (Hugginface Open-source models)
TestLLM (Experimental CPU-only)
What I’m Testing
I’m pushing the limits of small open-source models for AI network monitoring, specifically:
Function calling against live network services
How small can a model go while still handling:
Automated Nmap security scans
Quantum-readiness checks
Network Monitoring tasks
🟡 TestLLM – Current experimental model (llama.cpp on 2 CPU threads on huggingface docker space):
✅ Zero-configuration setup
⏳ 30s load time (slow inference but no API costs) . No token limited as the cost is low.
🔧 Help wanted! If you’re into edge-device AI, let’s collaborate!
Other Assistants
🟢 TurboLLM – Uses gpt-4.1-mini :
**It performs very well but unfortunatly OpenAI charges per token. For this reason tokens usage is limited.
Create custom cmd processors to run .net code on Quantum Network Monitor Agents
Real-time network diagnostics and monitoring
Security Audits
Penetration testing (Nmap/Metasploit)
🔵 HugLLM – Latest Open-source models:
🌐 Runs on Hugging Face Inference API. Performs pretty well using the lastest models hosted on Novita.
💡 Example commands you could test:
"Give me info on my websites SSL certificate"
"Check if my server is using quantum safe encyption for communication"
"Run a comprehensive security audit on my server"
'"Create a cmd processor to .. (what ever you want)" Note you need to install a Quantum Network Monitor Agent to run the .net code on. This is a very flexible and powerful feature. Use with caution!
Final Word
I fund the servers used to create these model files, run the Quantum Network Monitor service, and pay for inference from Novita and OpenAI—all out of my own pocket. All the code behind the model creation and the Quantum Network Monitor project is open source. Feel free to use whatever you find helpful.
If you appreciate the work, please consider buying me a coffee ☕. Your support helps cover service costs and allows me to raise token limits for everyone.
I'm also open to job opportunities or sponsorship.
