Modelpublic

SUPEROXIDES/HIVEMIND_PREVIEW_32B_-_GGUF

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face

6likes1.1kdownloads

Model Card

GGUF QUANTIZATIONS FOR HIVELABSAI'S "hivemind-32b-preview"

These quantizations were made with llama.cpp build 9181 using Bartowski's sample dataset for an importance matrix. The Q8_0, BF16, and F16 quantizations do not use any importance matrix.

As the original model is a preview release, these quantizations should be considered experimental, though functioning. I [SUPEROXIDES] am not endorsed or sponsored by Hive Labs.

<details> <summary>ORIGINAL README.MD FOR HIVEMIND PREVIEW 32B:</summary>

# Hivemind-32B-Preview

Hivemind-32B-Preview is a 32B-parameter model fine-tuned for multi-turn, emotionally attentive conversation in human-facing enterprise contexts. It is built on Qwen3-32B with a training set focused on conversational depth, emotional subtext, and sustained engagement across complex interpersonal scenarios.

Model Details

—Parameters: 32B
—Context length: 40,960 tokens
—Precision: bfloat16
—Base model: Qwen3-32B
—License: Proprietary, subject to upstream Qwen license terms

Training

Hivemind-32B-Preview was fine-tuned for multi-turn, human-facing conversations involving ambiguity and emotional subtext. The training set was purpose-built from enterprise interaction data.

Intended Use

Hivemind-32B-Preview is designed for enterprise human-agent partnership contexts: customer support, coaching-style interactions, and similar conversational deployments where sustained emotional attentiveness matters.

Scope and Ongoing Work

Hivemind-32B-Preview is a preview release. As with any conversational model, it has scope boundaries we are actively refining:

—It is not intended as a source of medical, legal, financial, or safety-critical advice, and should not be deployed in those contexts or as a replacement for professional human support.
—Performance is strongest in standard conversational scenarios.

We welcome failure-case reports from researchers and deployment partners at contact@hivelabs.dev.

Usage

vLLM (recommended)

bash

vllm serve HiveLabsAI/hivemind-32b-preview --dtype bfloat16

Transformers

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "HiveLabsAI/hivemind-32b-preview"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="bfloat16",
    device_map="auto",
)

messages = [{"role": "user", "content": "Your message here"}]
inputs = tokenizer.apply_chat_template(
    messages, return_tensors="pt", add_generation_prompt=True
).to(model.device)
outputs = model.generate(
    inputs, max_new_tokens=2048, temperature=0.6, top_p=0.95, top_k=20
)
print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True))

About

Hivemind is developed by Hive Labs. For research collaboration, deployment questions, or to report failure cases, contact contact@hivelabs.dev. </details>