ConeML 1.2B Experimental

ConeML 1.2B Experimental

Names the tallest, oldest or first item among facts you supply, as a small English research model in FP32 Safetensors and Q8_0 GGUF files. For researchers, non-commercial use only.

Made by ConeML on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: ConeML 1.2B Experimental is a small English research model. Given a set of stated relations, it answers which item is tallest, shortest, older, younger or first. The files come as FP32 Safetensors and as a Q8_0 GGUF. The maker calls it experimental and says coding and everyday chat are much weaker.

Who it suits: Researchers who study small models, supplied-fact reasoning and how answers change with wording or fact order.

Industry | Who uses it | How they use it | Example
AI research | Model evaluators | Test how a small model picks an endpoint from stated facts | Compare it with other small instruct models on shuffled facts
Academia | Students | Study how fact order and wording affect answers | Run the same question with ordered and shuffled facts
Open-source tooling | Hobbyists | Try a model with few active parameters | Load the GGUF file and ask a short relational question

Skip it if you want a chat assistant, a coding helper or anything for commercial use.

Where to use it:

• Python with the transformers library, for the Safetensors weights
• A local GGUF runner, for the Q8_0 file
• Windows, macOS or Linux machines that can fetch the files with Python

Use case | Who | What they get back
Endpoint questions | Researchers | The name of the highest, lowest or first item from your facts
Wording tests | Evaluators | Answers that show sensitivity to phrasing and fact order
Small-model comparison | Students | A reference point against other models of similar size

How to set it up: You need Python with pip and disk space for the files. The FP32 weights are about 4.7 GB, so check the repository's file list before you download everything.

• Python and pip
• Roughly 4.7 GB for FP32, less for the quantized file
• The transformers library or a GGUF runner

1. Install the Hugging Face hub tool.
2. Download the repository's files.
3. Load the Safetensors weights in transformers, or open the GGUF file in a runner.

pip install -U huggingface_hub
hf download ConeML/coneml-1.2b-experimental

How to use it: Write the facts in the prompt in any order, then ask which item sits at one end of the chain, such as the tallest or the earliest. The maker tested it with the model's own chat format.

hf download ConeML/coneml-1.2b-experimental

Tool | What it does
hf download | Copies this repository's files to your computer

Example: You list a few shuffled statements about who is taller than whom, then ask who is tallest. The model names one person. In the maker's tests at depth 5 it got 438 of 512 such questions right.

For newcomers and experts: If you are new to this, start with three or four short facts and one question, and check the answer by hand. Do not expect good results on code or open conversation.

If you know the field, the card describes about 348M active and 1.174B stored parameters. It compares the model with four small instruct models on shuffled endpoint questions. A later note says a simple set-difference rule solves every question, so the score shows endpoint selection, not multi-hop traversal.

Write the facts in any order
Ask for an endpoint
The model names one item
Check the answer yourself

Good to know: Success falls as chains get longer. The card lists 438 of 512 at depth 5, 240 at depth 12 and 74 at depth 24.

Title: Correct answers at depth 5, out of 512
ConeML: 438
Qwen2.5-1.5B-Instruct: 273
SmolLM2-1.7B-Instruct: 155
Llama-3.2-1B-Instruct: 136
Gemma-3-1B-IT: 138

• License: CC BY-NC 4.0
• Cost to run: none named
• Data it reaches: the files you download from Hugging Face, plus the text you type in
• Not for: commercial use, coding or everyday conversation, and proof of multi-hop reasoning

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 1.2B
• Weights: about 4.7 GB in F32; about 587 MB at 4-bit (estimate)
• Formats: Safetensors, GGUF
• Downloads in the last 30 days: 440
• Likes: 2
• License: cc-by-nc-4.0
• Languages: en
• Tags: transformers, safetensors, gguf, mixtral, text-generation, mixture-of-experts, experimental, conversational

Links

• Model: https://huggingface.co/ConeML/coneml-1.2b-experimental
• Files: https://huggingface.co/ConeML/coneml-1.2b-experimental/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download ConeML/coneml-1.2b-experimental

Then follow the usage in its model card. Nothing in this kit runs by itself.