Kasnol/Qwen3.6-35B-A3B-4bit
20.4 GB
Abr
Kasnol
AI & ML interests
None yet
Recent Activity
updated a bucket 7 days ago
Kasnol/Qwen3.6-35B-A3B-4bit published a bucket 7 days ago
Kasnol/Qwen3.6-35B-A3B-4bit updated a bucket 7 days ago
Kasnol/Qwen3.6-35B-A3B-4bit-DWQ-bucketOrganizations
None yet
updated a
bucket 7 days ago
published a
bucket 7 days ago
updated a
bucket 7 days ago
published a
bucket 7 days ago
updated a
bucket 8 days ago
published a
bucket 8 days ago
reacted to Anran-MLLM's post with š„ 2 months ago
Post
3583
š Introducing PerceptionDLM ā the first multimodal diffusion LLM for parallel region perception!
Most MLLMs are autoregressive, so captioning N regions costs N sequential passes. PerceptionDLM instead describes ALL masked regions in a single denoising process. š§©
āØ Highlights
ā¢ ā” Up to 3.4Ć faster on dense multi-region captioning, with stable per-image latency
ā¢ š PerceptionDLM-Base beats LLaDA-V on 15/16 multimodal benchmarks (new SOTA among open diffusion VLMs)
ā¢ š New benchmark: ParaDLC-Bench ā jointly evaluates caption quality AND inference efficiency
ā¢ š Code, models & benchmark all open-sourced
š¤ Models
MSALab/PerceptionDLM-Base
MSALab/PerceptionDLM
š Benchmark
MSALab/ParaDLC-Bench
š Paper: PerceptionDLM: Parallel Region Perception with Multimodal Diffusion Language Models (2606.19534)
š» Code: https://github.com/MSALab-PKU/PerceptionDLM
Diffusion LLMs aren't just for text ā they unlock efficient, parallel visual perception. šļøāØ
#multimodal #diffusion #VLM #perception
Most MLLMs are autoregressive, so captioning N regions costs N sequential passes. PerceptionDLM instead describes ALL masked regions in a single denoising process. š§©
āØ Highlights
ā¢ ā” Up to 3.4Ć faster on dense multi-region captioning, with stable per-image latency
ā¢ š PerceptionDLM-Base beats LLaDA-V on 15/16 multimodal benchmarks (new SOTA among open diffusion VLMs)
ā¢ š New benchmark: ParaDLC-Bench ā jointly evaluates caption quality AND inference efficiency
ā¢ š Code, models & benchmark all open-sourced
š¤ Models
MSALab/PerceptionDLM-Base
MSALab/PerceptionDLM
š Benchmark
MSALab/ParaDLC-Bench
š Paper: PerceptionDLM: Parallel Region Perception with Multimodal Diffusion Language Models (2606.19534)
š» Code: https://github.com/MSALab-PKU/PerceptionDLM
Diffusion LLMs aren't just for text ā they unlock efficient, parallel visual perception. šļøāØ
#multimodal #diffusion #VLM #perception
liked a
model 2 months ago
liked a
model 11 months ago
liked a
dataset over 1 year ago
reacted to Kseniase's post with š„ over 1 year ago
Post
3359
8 New Applications of Test-Time Scaling
We've noticed a huge interest in test-time scaling (TTS), so we decided to explore this concept further. Test-time compute (TTC) refers to the amount of computational power used by an AI model when generating a response. Many researchers are now focused on scaling TTC, as it enables slow, deep "thinking" and step-by-step reasoning, which improves overall models' performance.
Here are 8 fresh studies on test-time scaling:
1. Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach (2502.05171)
Introduces an LM that scales TTC by reasoning in latent space instead of generating more tokens with no special training. Here, a recurrent block to processes information iteratively.
2. Generating Symbolic World Models via Test-time Scaling of Large Language Models (2502.04728)
Shows how TTS is applied to enhance model's Planning Domain Definition Language (PDDL) reasoning capabilities, which can be used to generate a symbolic world model.
3. Can 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling (2502.06703)
Analyzes optimal TTS strategies and shows how small models can outperform much larger ones.
4. Llasa: Scaling Train-Time and Inference-Time Compute for Llama-based Speech Synthesis (2502.04128)
Shows how TTS improves expressiveness, timbre consistency and accuracy in speech synthesis with Llasa framework. It also dives into benefits of scaling train-time compute.
5. Rethinking Fine-Tuning when Scaling Test-Time Compute: Limiting Confidence Improves Mathematical Reasoning (2502.07154)
Suggests a modified training loss for better reasoning of LLMs when scaling TTC.
6. Adaptive Graph of Thoughts: Test-Time Adaptive Reasoning Unifying Chain, Tree, and Graph Structures (2502.05078)
Unifies the strengths of chain, tree, and graph paradigms into one framework that expands reasoning only on necessary subproblems.
7. Sample, Scrutinize and Scale: Effective Inference-Time Search by Scaling Verification (2502.01839)
Explores scaling trends of self-verification and how to improve its capabilities with TTC.
8. CodeMonkeys: Scaling Test-Time Compute for Software Engineering (2501.14723)
Explores how scaling serial compute (iterations) and parallel compute (trajectories), can improve accuracy in real-world software engineering issues.
Also, explore our article about TTS for more -> https://huggingface.co/blog/Kseniase/testtimecompute
We've noticed a huge interest in test-time scaling (TTS), so we decided to explore this concept further. Test-time compute (TTC) refers to the amount of computational power used by an AI model when generating a response. Many researchers are now focused on scaling TTC, as it enables slow, deep "thinking" and step-by-step reasoning, which improves overall models' performance.
Here are 8 fresh studies on test-time scaling:
1. Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach (2502.05171)
Introduces an LM that scales TTC by reasoning in latent space instead of generating more tokens with no special training. Here, a recurrent block to processes information iteratively.
2. Generating Symbolic World Models via Test-time Scaling of Large Language Models (2502.04728)
Shows how TTS is applied to enhance model's Planning Domain Definition Language (PDDL) reasoning capabilities, which can be used to generate a symbolic world model.
3. Can 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling (2502.06703)
Analyzes optimal TTS strategies and shows how small models can outperform much larger ones.
4. Llasa: Scaling Train-Time and Inference-Time Compute for Llama-based Speech Synthesis (2502.04128)
Shows how TTS improves expressiveness, timbre consistency and accuracy in speech synthesis with Llasa framework. It also dives into benefits of scaling train-time compute.
5. Rethinking Fine-Tuning when Scaling Test-Time Compute: Limiting Confidence Improves Mathematical Reasoning (2502.07154)
Suggests a modified training loss for better reasoning of LLMs when scaling TTC.
6. Adaptive Graph of Thoughts: Test-Time Adaptive Reasoning Unifying Chain, Tree, and Graph Structures (2502.05078)
Unifies the strengths of chain, tree, and graph paradigms into one framework that expands reasoning only on necessary subproblems.
7. Sample, Scrutinize and Scale: Effective Inference-Time Search by Scaling Verification (2502.01839)
Explores scaling trends of self-verification and how to improve its capabilities with TTC.
8. CodeMonkeys: Scaling Test-Time Compute for Software Engineering (2501.14723)
Explores how scaling serial compute (iterations) and parallel compute (trajectories), can improve accuracy in real-world software engineering issues.
Also, explore our article about TTS for more -> https://huggingface.co/blog/Kseniase/testtimecompute
- 1 reply
reacted to prithivMLmods's post with š almost 2 years ago
Post
4055
GRID-6X : Layout for Seamless Image Assembly š„
šŖØGit: github.com/PRITHIVSAKTHIUR/GRID-6X
šŖØDoc / Blog: https://huggingface.co/blog/prithivMLmods/grid-6x
In the
1. **Image Initialization**:
2. **Image Placement**: Images are pasted onto the canvas using a loop:
- **Horizontal Position**:
- **Vertical Position**:
1. **Grid Size Selection**: The user selects the grid size from options like "2x1", "1x2", "2x2", "2x3", "3x2", and "1x1". Each option corresponds to the arrangement of images:
- **2x1**: 2 images in a single row
- **1x2**: 1 image in two rows (column layout)
- **2x2**: 2 rows with 2 images each
- **2x3**: 2 rows with 3 images each
- **3x2**: 3 rows with 2 images each
- **1x1**: A single image (default)
2. **Image Generation**: Based on the grid size selection, the app calculates the number of images to generate. For example:
- If the grid size is "2x2", the app generates 4 images.
- For "3x2", it generates 6 images.
-> Each option arranges images accordingly, providing flexibility in viewing multiple images in one output.
-> Added both of these spaces that support the GRID functionality Layout for Seamless Image Assembly :
----------
š„IMAGINEO-4K: https://huggingface.co/spaces/prithivMLmods/IMAGINEO-4K
š„GRID-6X: https://huggingface.co/spaces/prithivMLmods/GRID-6X
----------
.
.
.@prithivMLmods š¤
šŖØGit: github.com/PRITHIVSAKTHIUR/GRID-6X
šŖØDoc / Blog: https://huggingface.co/blog/prithivMLmods/grid-6x
In the
infer function:grid_img = Image.new('RGB', (width * grid_size_x, height * grid_size_y))
for i, img in enumerate(result.images[:num_images]):
grid_img.paste(img, (i % grid_size_x * width, i // grid_size_x * height))1. **Image Initialization**:
grid_img is a blank canvas that will hold the images in a grid format.2. **Image Placement**: Images are pasted onto the canvas using a loop:
- **Horizontal Position**:
(i % grid_size_x) * width calculates the x-coordinate.- **Vertical Position**:
(i // grid_size_x) * height calculates the y-coordinate.1. **Grid Size Selection**: The user selects the grid size from options like "2x1", "1x2", "2x2", "2x3", "3x2", and "1x1". Each option corresponds to the arrangement of images:
- **2x1**: 2 images in a single row
- **1x2**: 1 image in two rows (column layout)
- **2x2**: 2 rows with 2 images each
- **2x3**: 2 rows with 3 images each
- **3x2**: 3 rows with 2 images each
- **1x1**: A single image (default)
2. **Image Generation**: Based on the grid size selection, the app calculates the number of images to generate. For example:
- If the grid size is "2x2", the app generates 4 images.
- For "3x2", it generates 6 images.
-> Each option arranges images accordingly, providing flexibility in viewing multiple images in one output.
-> Added both of these spaces that support the GRID functionality Layout for Seamless Image Assembly :
----------
š„IMAGINEO-4K: https://huggingface.co/spaces/prithivMLmods/IMAGINEO-4K
š„GRID-6X: https://huggingface.co/spaces/prithivMLmods/GRID-6X
----------
.
.
.@prithivMLmods š¤
reacted to yongchanghao's post with š„ almost 2 years ago
Post
3850
We just released a paper (NeuZip) that compresses VRAM in a lossless manner to run larger models. This should be particularly useful when VRAM is insufficient during training/inference. Specifically, we look inside each floating number and find that the exponents are highly compressible (as shown in the figure below).
Read more about the work at NeuZip: Memory-Efficient Training and Inference with Dynamic Compression of Neural Networks (2410.20650)
Read more about the work at NeuZip: Memory-Efficient Training and Inference with Dynamic Compression of Neural Networks (2410.20650)
liked a
Space almost 2 years ago
liked a
model almost 2 years ago
updated a
model almost 2 years ago
liked a
dataset almost 2 years ago
reacted to davidberenstein1957's post with š almost 2 years ago
Post
1462
Want to get familiar with llama 3.2 and an actual example of structured data generation? Try out the Google Colab I created.
Notebook: https://colab.research.google.com/drive/1nHNXUbgwRMyjFeBZbQvNLqaXGkqE4Wcs?usp=sharing
Model: hugging-quants/Llama-3.2-3B-Instruct-Q8_0-GGUF
Dataset: argilla/FinePersonas-v0.1
Library: https://github.com/argilla-io/distilabel
Notebook: https://colab.research.google.com/drive/1nHNXUbgwRMyjFeBZbQvNLqaXGkqE4Wcs?usp=sharing
Model: hugging-quants/Llama-3.2-3B-Instruct-Q8_0-GGUF
Dataset: argilla/FinePersonas-v0.1
Library: https://github.com/argilla-io/distilabel
liked 2
models almost 2 years ago
