Downloads · 30 days
6
3% of all-time downloads
vidhyakshayakannan/appointment-intent-llama32-1b-sft-GGUF
appointment-intent-llama32-1b-sft-GGUF is a machine learning model from vidhyakshayakannan. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for gguf.
This repository contains an Ollama-ready GGUF export of a fine-tuned Llama 3.2 1B Instruct model for appointment-booking intent classification.
Downloads · 30 days
6
3% of all-time downloads
All-time downloads
178
Public
Repo size
808 MB
Likes
0
Public
Click a slice to open those files.
.gguf808 MB · 100%
From the Hugging Face model README
This repository contains an Ollama-ready GGUF export of a fine-tuned Llama 3.2 1B Instruct model for appointment-booking intent classification.
The model is designed to act as the intent-routing layer for a scheduling agent. It reads a user message and returns exactly one intent label, which an application can map to a scheduling workflow such as listing appointments, cancelling an appointment, rescheduling an appointment, or updating provider availability.
Scheduling agents often need a reliable routing step before they call tools or APIs. A user may say:
Can you move my appointment from Monday to Friday?
The scheduling agent first classifies the message:
RESCHEDULE_APPOINTMENTS
The application can then route the request to a rescheduling flow, collect any missing details, call the calendar or appointment API, and ask for confirmation before making changes.
This model is intentionally narrow: it is a classifier, not a general-purpose assistant. It should return a single label and no explanation.
LIST_APPOINTMENTS
CANCEL_APPOINTMENTS
RESCHEDULE_APPOINTMENTS
UNSUPPORTED_REQUEST
BLOCK_SLOTS
UNBLOCK_SLOTS
UPDATE_SPECIAL_SLOTS
CONVERSATIONAL_GREETING
CLOSING_CONVERSATION
The fine-tuning notebook follows this process:
USER_INPUT and TARGET_INTENT.meta-llama/Llama-3.2-1B-Instruct model on the classification task.system: intent-classification instructions
user: the scheduling message
assistant: the correct intent label
Q4_K_M for local Ollama usage.meta-llama/Llama-3.2-1B-Instructvidhyakshayakannan/appointment-intent-llama32-1b-sftQ4_K_MThe dataset contains 301 labeled examples with three columns:
TEST_ID: row identifierUSER_INPUT: user messageTARGET_INTENT: correct intent labelClass distribution:
| Intent | Examples |
|---|---|
BLOCK_SLOTS | 61 |
RESCHEDULE_APPOINTMENTS | 51 |
LIST_APPOINTMENTS | 46 |
UNBLOCK_SLOTS | 39 |
UPDATE_SPECIAL_SLOTS | 35 |
CANCEL_APPOINTMENTS | 32 |
UNSUPPORTED_REQUEST | 25 |
CONVERSATIONAL_GREETING | 6 |
CLOSING_CONVERSATION | 6 |
The fine-tuning run used an 80/20 stratified split: 240 training examples and 61 held-out test examples.
Held-out test results from the fine-tuning notebook:
| Model | Test Accuracy | Macro F1 | Weighted F1 |
|---|---|---|---|
| Base Llama 3.2 1B Instruct | 39.3% | 0.1753 | 0.3134 |
| Fine-tuned model | 85.2% | 0.8784 | 0.8549 |
The accuracy improvement shows that the model learned the appointment-intent classification format. For production use, evaluate on a larger, more representative dataset that includes real user phrasing, ambiguous requests, spelling variation, and out-of-distribution inputs.
The recommended local setup is to create an Ollama alias using the included Modelfile. The alias includes the same system instruction and chat template used during fine-tuning.
Download and install Ollama:
https://ollama.com/download
Confirm installation:
ollama --version
ollama pull hf.co/vidhyakshayakannan/appointment-intent-llama32-1b-sft-GGUF:Q4_K_M
curl -L https://huggingface.co/vidhyakshayakannan/appointment-intent-llama32-1b-sft-GGUF/resolve/main/Modelfile -o Modelfile
ollama create appointment-intent-sft -f Modelfile
ollama run appointment-intent-sft "I need to move my appointment from Monday to Friday."
Expected output:
RESCHEDULE_APPOINTMENTS
More examples:
ollama run appointment-intent-sft "Can you cancel my appointment tomorrow?"
ollama run appointment-intent-sft "What appointments do I have next week?"
ollama run appointment-intent-sft "Please block off my calendar from 2 to 4 PM."
ollama run appointment-intent-sft "Hello, good morning."
Expected labels:
CANCEL_APPOINTMENTS
LIST_APPOINTMENTS
BLOCK_SLOTS
CONVERSATIONAL_GREETING
You can run the GGUF directly:
ollama run hf.co/vidhyakshayakannan/appointment-intent-llama32-1b-sft-GGUF:Q4_K_M "What appointments do I have next week?"
For the most consistent classifier behavior, use the appointment-intent-sft alias created from the included Modelfile.
Input should be a natural-language scheduling message.
Output should be exactly one intent label and nothing else.
Good input:
Can you move my appointment to next Friday?
Good output:
RESCHEDULE_APPOINTMENTS
Because the model was fine-tuned as a classifier, applications should use it for routing rather than asking it for conversational explanations.
A simple scheduling-agent architecture can use this model as the first step:
UNSUPPORTED_REQUEST and ambiguous messages to a fallback flow or human review.Example intent-to-tool mapping:
| Intent | Possible Application Action |
|---|---|
LIST_APPOINTMENTS | Query upcoming appointments |
CANCEL_APPOINTMENTS | Start cancellation flow |
RESCHEDULE_APPOINTMENTS | Start rescheduling flow |
BLOCK_SLOTS | Block provider availability |
UNBLOCK_SLOTS | Remove blocked availability |
UPDATE_SPECIAL_SLOTS | Update exception hours or special slots |
CONVERSATIONAL_GREETING | Send greeting response |
CLOSING_CONVERSATION | End or summarize conversation |
UNSUPPORTED_REQUEST | Fallback or human handoff |
If ollama pull hf.co/... fails, update Ollama to a recent version that supports pulling GGUF models directly from Hugging Face.
If the model returns explanations or unexpected text, use the appointment-intent-sft alias created from the included Modelfile instead of running the raw GGUF directly.
If the model returns the wrong label for an obvious example, confirm that temperature is set to 0 and run:
ollama run appointment-intent-sft "your message here"
If you receive a 404, copy the model path exactly:
hf.co/vidhyakshayakannan/appointment-intent-llama32-1b-sft-GGUF:Q4_K_M
This is a compact model trained on a small labeled dataset for a narrow scheduling-intent task. It is not a medical assistant and should not provide medical advice. Production deployment should include: