MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic

MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic

FP8 dynamic quantized version of MamayLM-Gemma-3-27B-IT-v2.0. MamayLM v2.0 is a series of Ukrainian-adapted LLMs based on Gemma 3, developed by INSAIT.

Blog post: MamayLM v2.0 Release

Key improvements over MamayLM v1.0

1. Vision-language understanding — The model understands both text and images within the same context.
2. Instruction-following — Trained on a broader range of tasks, multi-turn conversations, complex instructions, and system prompts.
3. Longer context — Effective context of 131k tokens for longer conversations and complex instructions.
4. Updated knowledge cut-off — Pretraining data up to May 2025, instruction fine-tuning up to October 2025.

• Quantization: FP8 dynamic (weights FP8_E4M3 per-tensor, activations quantized per-token at runtime)
• Quantized layers: All linear layers in the language model
• Full precision: Vision tower, multimodal projector, lm_head
• Hardware: Requires a GPU with compute capability ≥ 8.9 (H100, H200, RTX 4090)

Note: Do not load this model with vLLM's --quantization fp8 flag — the checkpoint is already quantized and identifies itself via config.json. vLLM's on-the-fly FP8 flag is broken for multimodal Gemma 3 (it silently quantizes the vision tower and produces garbage image outputs); this pre-quantized checkpoint avoids that bug.

Usage

vLLM

from vllm import LLM, SamplingParams

llm = LLM(model="INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic")
params = SamplingParams(max_tokens=512, temperature=0.2)

messages = [
    {
        "role": "user",
        "content": [{"type": "text", "text": "Коли був заснований Київський університет?"}],
    },
]

outputs = llm.chat(messages, sampling_params=params)
print(outputs[0].outputs[0].text)

Serving with the OpenAI-compatible API:

vllm serve INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic

Transformers

from transformers import AutoProcessor, Gemma3ForConditionalGeneration
import torch

model_id = "INSAIT-Institute/MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic"

processor = AutoProcessor.from_pretrained(model_id)
model = Gemma3ForConditionalGeneration.from_pretrained(
    model_id, device_map="auto"
).eval()

messages = [
    {
        "role": "user",
        "content": [{"type": "text", "text": "Коли був заснований Київський університет?"}],
    },
]

inputs = processor.apply_chat_template(
    messages, add_generation_prompt=True, tokenize=True,
    return_dict=True, return_tensors="pt"
).to(model.device, dtype=torch.bfloat16)

input_len = inputs["input_ids"].shape[-1]

with torch.inference_mode():
    generation = model.generate(**inputs, max_new_tokens=512, do_sample=True, temperature=0.2)
    generation = generation[0][input_len:]

print(processor.decode(generation, skip_special_tokens=True))

With an image

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Опиши, що ти бачиш на зображенні."},
        ],
    },
]

inputs = processor.apply_chat_template(
    messages, add_generation_prompt=True, tokenize=True,
    return_dict=True, return_tensors="pt"
).to(model.device, dtype=torch.bfloat16)

input_len = inputs["input_ids"].shape[-1]

with torch.inference_mode():
    generation = model.generate(**inputs, max_new_tokens=512, do_sample=True, temperature=0.2)
    generation = generation[0][input_len:]

print(processor.decode(generation, skip_special_tokens=True))

License

MamayLM-Gemma-3-27B-IT-v2.0-FP8-dynamic is distributed under the Gemma Terms of Use.