Skip to content
Other

LLaVA-1.5 13B

LoRA-адаптер ICVCO с random-негативами

1 min readupdated

Это PEFT LoRA-адаптер для llava-hf/llava-1.5-13b-hf, обученный с целью ICVCO на варианте данных random. Автор — pratikjalan. Это часть сетки из 60 запусков (5 целей × 3 варианта × 4 базовые модели), где проверяют, превосходят ли SelfPO-негативы случайные. Веса базовой модели в репозитории нет.

Ключевые особенности

  • PEFT LoRA поверх LLaVA-1.5-13B: ранг 64, alpha 128, dropout 0.0.
  • Обучены только слои языковой модели; vision tower и projector заморожены.
  • Вариант random — равномерно случайный отбор негативов, контролируемый baseline.
  • 1 эпоха, lr 5e-06, cosine/warmup 0.1, global batch 64, bfloat16 без квантизации.
  • 304 шага; loss снизился с 2.6501 до 1.3192.

Пример

from peft import PeftModel
from transformers import AutoProcessor, LlavaForConditionalGeneration
base = LlavaForConditionalGeneration.from_pretrained("llava-hf/llava-1.5-13b-hf", torch_dtype="bfloat16")
model = PeftModel.from_pretrained(base, "pratikjalan/selfpo-llava15-13b-icvco-random")
processor = AutoProcessor.from_pretrained("llava-hf/llava-1.5-13b-hf")
ПодробнееLoss: 2.6501 → 1.3192 за 304 шага.
LoRA: rank 64, alpha 128, только LM-слои; cutoff 4096, beta 0.1.
Лицензия CC-BY-NC-4.0, некоммерческая; базовые веса не входят.
Это исследовательский артефакт, не safety-tuned и не production-модель: возможны галлюцинации и смещения данных. Бенчмарков в карточке нет.
Discuss Telegram 0 9

Will it run?

Runs in FP8 / INT8

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • F16weights 26 GB · ready files29 GB
  • FP8 / INT8recommendedweights 13 GB15 GB
  • 4-bitweights 7.9 GB9.2 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 5080 · F16 — from RTX 5090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model llava-1.5-13b-hf.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # pratikjalan/selfpo-llava15-13b-icvco-random
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download llava-hf/llava-1.5-13b-hf --local-dir ~/models/llava-1.5-13b-hf
    hf download pratikjalan/selfpo-llava15-13b-icvco-random --local-dir ~/models/selfpo-llava15-13b-icvco-random
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to llava-1.5-13b-hf
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

llava-hf/llava-1.5-13b-hf

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of llava-1.5-13b-hf (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 2All on Hugging Face
Fine-tunes 5All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

pratikjalan/selfpo-llava15-13b-icvco-random
Author
pratikjalan
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

90

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments