Skip to content
Other

Time-R1-7B LoRA: Video Evidence Auditing

2 min readupdated

This LoRA adapter (a lightweight add-on for the base model) is fine-tuned to improve accuracy and reliance on actual evidence in video temporal grounding tasks. Built on top of Time-R1-7B and developed by the authors of the Accuracy Is Not Evidence study, it helps pinpoint event time intervals in videos more precisely using placebo-controlled training. The model is ideal for researchers and developers working on video analysis and reliability audits for multimodal models.

Key features

  • Placebo control: Training with a loss function that controls answer validity on altered videos.
  • Video processing: Support for video inputs at 2 fps and lengths up to 768 frames.
  • LoRA parameters: Rank r=16, alpha=32, dropout 0.05, applied to q/k/v/o and gate/up/down projections.
  • Training: 300 optimizer steps on 2,400 pairs from the Charades-STA dataset.

What you can do

  • Run evaluation: Test the model using audit scripts from the official research repository.
  • Integrate into pipeline: Load the adapter via the PEFT library for video event analysis.
  • Conduct an audit: Use the model to measure how much a multimodal system's answers depend on the video sequence.

Example

# pip install "transformers>=5.12" "peft>=0.19.1" qwen-vl-utils
import torch
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
from peft import PeftModel

base = "Boshenxx/Time-R1-7B"
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(base, torch_dtype=torch.bfloat16, device_map="auto")
processor = AutoProcessor.from_pretrained(base)
model = PeftModel.from_pretrained(model, "jinzijie-bupt/evidence-audit-timer1-7b-pc-a05-s0")
Read moreDetails
Alpha coefficient (α): 0.5
Beta coefficient (β): 0.1
Learning rate: 5e-6
Requirements: transformers 5.x (tested on 5.12.1), peft 0.19.1 or newer
Discuss Telegram 0 8

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 16 GB · ready files18 GB
  • FP8 / INT8recommendedweights 7.9 GB9.2 GB
  • 4-bitweights 4.5 GB5.5 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 3080 · BF16 — from RTX 4090

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 Time-R1-7B.

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.

    # jinzijie-bupt/evidence-audit-timer1-7b-pc-a05-s0
    # ===== 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 Boshenxx/Time-R1-7B --local-dir ~/models/Time-R1-7B
    hf download jinzijie-bupt/evidence-audit-timer1-7b-pc-a05-s0 --local-dir ~/models/evidence-audit-timer1-7b-pc-a05-s0
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Time-R1-7B
  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

Boshenxx/Time-R1-7B

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

New posts about versions of Time-R1-7B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 2All on Hugging Face
Fine-tunes 1All on Hugging Face
LoRAs and adapters 5All on Hugging Face

About the model

jinzijie-bupt/evidence-audit-timer1-7b-pc-a05-s0
Author
jinzijie-bupt
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

80

Ask AI about this model

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

Comments