dharavathsunil

geoai-outdoor-slm

Deploy Dedicated

README

License: apache-2.0

Base model and changes

Base model: Qwen/Qwen2.5-1.5B-Instruct (1.5B parameters, Apache-2.0).
Changes: none. The weights in this repository are an unmodified copy of the base model, with no fine-tuning.
qwen2.5-1.5b-instruct-q4_k_m.gguf is the official 4-bit quantisation from Qwen/Qwen2.5-1.5B-Instruct-GGUF. It is the file the live assistant runs on CPU with llama.cpp.

How the model is used

The assistant answers "where and when should I go outdoors" questions. The model has two narrow jobs:

Parsing a free-form question into JSON (activity, place, radius, profile) with schema-constrained decoding. Dates and times are resolved by plain code, and the place is checked against the question text.
Phrasing a 2-3 sentence summary from facts that were fetched live (weather, air quality, OpenStreetMap context). Every number and place name in the output is verified against those facts. If a check fails, a template sentence replaces the model text.

All recommendations, rankings and numbers come from live data and rule-based scoring, not from the model's memory.

Intended use and limitations

Intended for outdoor-activity advice in India as part of the full pipeline. Not a general-purpose assistant.
Not medical advice. Do not use it alone for health decisions.
A 1.5B model can phrase things awkwardly. That is why facts and numbers are rendered by code.

Load it

python

from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("dharavathsunil/geoai-outdoor-slm")model = AutoModelForCausalLM.from_pretrained("dharavathsunil/geoai-outdoor-slm")

python

from huggingface_hub import hf_hub_downloadfrom llama_cpp import Llamapath = hf_hub_download("dharavathsunil/geoai-outdoor-slm", "qwen2.5-1.5b-instruct-q4_k_m.gguf")llm = Llama(model_path=path, n_ctx=4096)

Data and credits

Map data © OpenStreetMap contributors (Overpass API, Nominatim). Weather and air quality: Open-Meteo. Base model by the Qwen team, Alibaba Cloud (Apache-2.0). Code written with the help of an AI coding assistant (Claude), as disclosed in the project README.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

dharavathsunil

Model Tree

Base

Qwen/Qwen2.5-1.5B-Instruct

Quantized

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Base model and changes

Base model: Qwen/Qwen2.5-1.5B-Instruct (1.5B parameters, Apache-2.0).
Changes: none. The weights in this repository are an unmodified copy of the base model, with no fine-tuning.
qwen2.5-1.5b-instruct-q4_k_m.gguf is the official 4-bit quantisation from Qwen/Qwen2.5-1.5B-Instruct-GGUF. It is the file the live assistant runs on CPU with llama.cpp.

How the model is used

The assistant answers "where and when should I go outdoors" questions. The model has two narrow jobs:

Parsing a free-form question into JSON (activity, place, radius, profile) with schema-constrained decoding. Dates and times are resolved by plain code, and the place is checked against the question text.
Phrasing a 2-3 sentence summary from facts that were fetched live (weather, air quality, OpenStreetMap context). Every number and place name in the output is verified against those facts. If a check fails, a template sentence replaces the model text.

All recommendations, rankings and numbers come from live data and rule-based scoring, not from the model's memory.

Intended use and limitations

Intended for outdoor-activity advice in India as part of the full pipeline. Not a general-purpose assistant.
Not medical advice. Do not use it alone for health decisions.
A 1.5B model can phrase things awkwardly. That is why facts and numbers are rendered by code.

Load it

python

from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("dharavathsunil/geoai-outdoor-slm")model = AutoModelForCausalLM.from_pretrained("dharavathsunil/geoai-outdoor-slm")

python

from huggingface_hub import hf_hub_downloadfrom llama_cpp import Llamapath = hf_hub_download("dharavathsunil/geoai-outdoor-slm", "qwen2.5-1.5b-instruct-q4_k_m.gguf")llm = Llama(model_path=path, n_ctx=4096)