---

base_model: ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM

tags:

• gguf
• llama.cpp
• quantized
• qwen3
• function-calling

license: apache-2.0

language:

• en

datasets:

• Salesforce/xlam-function-calling-60k

pipeline_tag: text-generation

---

Qwen3.5-0.8B — Function Calling (xLAM) — GGUF

GGUF quantizations of ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM,

a function-calling fine-tune trained on Salesforce's xLAM dataset and converted with

llama.cpp.

| Field | Value |

|---|---|

| Source checkpoint | ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM |

| Base model | Qwen/Qwen3.5-0.8B |

| Dataset | Salesforce/xlam-function-calling-60k |

| Objective | Supervised fine-tuning for structured tool / function calling |

| Quantization tool | llama.cpp convert_hf_to_gguf.py + llama-quantize |

> Conversion note. The published source repo stores bitsandbytes 4-bit (nf4) weights, which llama.cpp cannot convert directly. These GGUFs were therefore built from the full-precision merged checkpoint produced during training, so they reflect the model's fp16 weights rather than a re-quantized 4-bit copy.

Available quantizations

| File | Size | Notes |

|---|---|---|

| qwen3.5-0.8b-function-calling-xlam.q4_k_m.gguf | 529 MB (recommended) | 4-bit K-quant medium; best size/quality balance |

| qwen3.5-0.8b-function-calling-xlam.q5_k_m.gguf | 578 MB (balanced) | 5-bit K-quant medium; near-full quality |

| qwen3.5-0.8b-function-calling-xlam.q8_0.gguf | 812 MB (largest) | 8-bit; closest to the source precision |

Recommended default: Q4_K_M. For maximum fidelity use Q8_0; for the smallest

footprint use Q4_K_M (no lower-bit quants are published for this model).

Usage

llama.cpp

# One-shot
llama-cli -hf ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM-GGUF --jinja -p "Your prompt here" -n 256

# Interactive chat
llama-cli -hf ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM-GGUF --jinja -cnv

Ollama

ollama run hf.co/ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM-GGUF:Q4_K_M

llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM-GGUF",
    filename="*q4_k_m.gguf",
    n_ctx=4096,
)
out = llm.create_chat_completion(
    messages=[{"role": "user", "content": "Your prompt here"}],
    max_tokens=256,
)
print(out["choices"][0]["message"]["content"])

Intended use

Research and non-commercial experimentation with structured function / tool calling.

Prompt the model with a user request plus the available tool schemas; it emits a

function call. Verify all emitted calls before executing them downstream.

Limitations

• GGUF quantizations carry unavoidable quality loss relative to the source weights;

prefer Q8_0 when fidelity matters.

• Inherits every limitation of the source checkpoint

(ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM).

• Function-calling behaviour is shaped by the 60 schema families in the training set;

performance on out-of-distribution APIs may degrade.

Citation

@misc{qwen3_5_0_8b_gguf,
  author       = {Ermia Azarkhalili},
  title        = {Qwen3.5-0.8B — Function Calling (xLAM) — GGUF quantized},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/ermiaazarkhalili/Qwen3.5-0.8B-Function-Calling-xLAM-GGUF}}
}