Article
An Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instruct
By
•
•
63Austin Liu
Austin362667
AI & ML interests
None yet
Recent Activity
upvoted
an
article
2 days ago
An Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instruct
upvoted
an
article
6 days ago
Parquet Content-Defined Chunking
upvoted
an
article
7 days ago
SmolVLA: Efficient Vision-Language-Action Model trained on Lerobot Community Data
Organizations
None yet
upvoted
an
article
2 days ago
upvoted
an
article
6 days ago
Article
Parquet Content-Defined Chunking
By
•
•
50
upvoted
an
article
7 days ago
Article
SmolVLA: Efficient Vision-Language-Action Model trained on Lerobot Community Data
By
and 8 others
•
•
213
reacted to
danaaubakirova's
post with 🤗❤️
7 days ago
Post
2363
We just dropped SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics!
check out the blog: https://huggingface.co/blog/smolvla
read the technical report: SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics (2506.01844)
access the model weights: lerobot/smolvla_base
check out the blog: https://huggingface.co/blog/smolvla
read the technical report: SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics (2506.01844)
access the model weights: lerobot/smolvla_base
reacted to
andito's
post with 👀🔥
7 days ago
Post
3960
🧠👁️ Can AI visualize solutions?
Humans often solve visual problems by sketching ideas in our minds. What if Vision-Language Models (VLMs) could do something similar, not by generating full images, but by using internal “mental sketches”?
That’s the idea behind Mirage, a new framework that empowers VLMs to reason using latent visual tokens. Instead of just thinking in words, Mirage mixes in abstract visual representations that help the model solve complex tasks.
These aren't photorealistic images. They're compact, internal representations optimized purely to support reasoning.
🔧 Mirage is trained in two phases:
1) Grounding: It learns to produce latent tokens anchored in real images.
2) Refinement: The model drops the images and learns to generate visual tokens on its own.
📈 And yes, it works!
On challenging benchmarks like Visual Spatial Planning, Jigsaw puzzles, and Spatial Attention Tasks, Mirage clearly outperforms GPT-4o and other strong baselines.
Smart sketches > empty words.
By mimicking the way humans visualize solutions, Mirage gives AI a new kind of imagination, one that’s faster, more efficient, and more human-like.
Kudos to the teams at UMass Amherst and MIT behind this exciting work.
Check the paper: Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens (2506.17218)
Humans often solve visual problems by sketching ideas in our minds. What if Vision-Language Models (VLMs) could do something similar, not by generating full images, but by using internal “mental sketches”?
That’s the idea behind Mirage, a new framework that empowers VLMs to reason using latent visual tokens. Instead of just thinking in words, Mirage mixes in abstract visual representations that help the model solve complex tasks.
These aren't photorealistic images. They're compact, internal representations optimized purely to support reasoning.
🔧 Mirage is trained in two phases:
1) Grounding: It learns to produce latent tokens anchored in real images.
2) Refinement: The model drops the images and learns to generate visual tokens on its own.
📈 And yes, it works!
On challenging benchmarks like Visual Spatial Planning, Jigsaw puzzles, and Spatial Attention Tasks, Mirage clearly outperforms GPT-4o and other strong baselines.
Smart sketches > empty words.
By mimicking the way humans visualize solutions, Mirage gives AI a new kind of imagination, one that’s faster, more efficient, and more human-like.
Kudos to the teams at UMass Amherst and MIT behind this exciting work.
Check the paper: Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens (2506.17218)
- 4 replies
reacted to
eliebak's
post with 👀🔥
7 days ago
Post
4455
Kimi K2 tech report is full of gems as always. Here are my notes on it:
> MuonClip: Pretty crazy how after 70k the training stabilizes and the QK-clip is basically inactive. There is also no loss in perf with QK-clip which is not trivial at all (at small scale but with aggressive threshold). Also a cool explanation of why muon makes the logit explode in appendix E (tl;dr is that muon makes the singular value of the update matrix higher)
> Sparsity scaling laws to justify their ratio, they have a very solid training infra that allows the model to be trained at this sparsity level, they could have increased even more but as sparsity increases the training becomes less efficient.
> They diminish the number of attention heads to make it more efficient for long context since attention heads are a big bottleneck for long context. They also remove 2 of the 3 "first dense" layers in the dsv3 arch.
With the sparsity and attention heads (divided by 2) they achieve 83% increased flops compared to deepseek v3 arch at 128k.
> Data: Rephrasing is KEY. They do a lot more synthetic data generation and rephrase their corpus to have different styles, for longer documents they do it by chunk. I'm (half) surprised by the fact that ONLY 1 epoch (assuming same number of training tokens I think?) of data rephrased 10 times has better accuracy than 10 epochs of the same data rephrased once.
> They do rewriting for Math and Knowledge, for Math they apply the ShallowMath recipe and instruct the model to rephrase in a "learning note" style
> They talk about diversity and probably have some internal stuff/eval to test that, as always still a bit unclear for me how to properly measure that.
The infra is also very nice, quick summary:
> PP=16 (1F1B schedule, a bit custom), EP=16, zero1
> No FP8 computation but for storage of specific layers, selective recomputation for inexpensive block, activation offloading to CPU
> MuonClip: Pretty crazy how after 70k the training stabilizes and the QK-clip is basically inactive. There is also no loss in perf with QK-clip which is not trivial at all (at small scale but with aggressive threshold). Also a cool explanation of why muon makes the logit explode in appendix E (tl;dr is that muon makes the singular value of the update matrix higher)
> Sparsity scaling laws to justify their ratio, they have a very solid training infra that allows the model to be trained at this sparsity level, they could have increased even more but as sparsity increases the training becomes less efficient.
> They diminish the number of attention heads to make it more efficient for long context since attention heads are a big bottleneck for long context. They also remove 2 of the 3 "first dense" layers in the dsv3 arch.
With the sparsity and attention heads (divided by 2) they achieve 83% increased flops compared to deepseek v3 arch at 128k.
> Data: Rephrasing is KEY. They do a lot more synthetic data generation and rephrase their corpus to have different styles, for longer documents they do it by chunk. I'm (half) surprised by the fact that ONLY 1 epoch (assuming same number of training tokens I think?) of data rephrased 10 times has better accuracy than 10 epochs of the same data rephrased once.
> They do rewriting for Math and Knowledge, for Math they apply the ShallowMath recipe and instruct the model to rephrase in a "learning note" style
> They talk about diversity and probably have some internal stuff/eval to test that, as always still a bit unclear for me how to properly measure that.
The infra is also very nice, quick summary:
> PP=16 (1F1B schedule, a bit custom), EP=16, zero1
> No FP8 computation but for storage of specific layers, selective recomputation for inexpensive block, activation offloading to CPU
reacted to
andito's
post with ❤️🔥
7 days ago
Post
2671
Many VLMs claim to process hours of video. But can they follow the story?🤔
Today, we introduce TimeScope: The benchmark that separates true temporal understanding from marketing hype. Let's see how much VLMs really understand!⏳
We test three skills that matter for real-world use:
🔎 Localized Retrieval: Find a specific action.
🧩 Information Synthesis: Piece together scattered clues.
🏃 Fine-Grained Perception: Analyze detailed motion (e.g., count how many times a person swings an axe).
The results are in, and they're revealing. Only Gemini 2.5 pro handles 1-hour-long videos.
Performance drops sharply with duration, proving that long video understanding is still challenging. We've found the breaking points—now the community can start fixing them.📈
Want to learn more? TimeScope is 100% open-source. Benchmark your model and help us build the next generation of video AI.
📖 Blog:
https://huggingface.co/blog/timescope-video-lmm-benchmark
👩💻 Leaderboard & Demo: Apollo-LMMs/TimeScope
📊 Dataset: Apollo-LMMs/TimeScope
⚙️ Eval Code: https://github.com/EvolvingLMMs-Lab/lmms-eval
Today, we introduce TimeScope: The benchmark that separates true temporal understanding from marketing hype. Let's see how much VLMs really understand!⏳
We test three skills that matter for real-world use:
🔎 Localized Retrieval: Find a specific action.
🧩 Information Synthesis: Piece together scattered clues.
🏃 Fine-Grained Perception: Analyze detailed motion (e.g., count how many times a person swings an axe).
The results are in, and they're revealing. Only Gemini 2.5 pro handles 1-hour-long videos.
Performance drops sharply with duration, proving that long video understanding is still challenging. We've found the breaking points—now the community can start fixing them.📈
Want to learn more? TimeScope is 100% open-source. Benchmark your model and help us build the next generation of video AI.
📖 Blog:
https://huggingface.co/blog/timescope-video-lmm-benchmark
👩💻 Leaderboard & Demo: Apollo-LMMs/TimeScope
📊 Dataset: Apollo-LMMs/TimeScope
⚙️ Eval Code: https://github.com/EvolvingLMMs-Lab/lmms-eval
upvoted
an
article
7 days ago
Article
TimeScope: How Long Can Your Video Large Multimodal Model Go?
By
and 3 others
•
•
31
upvoted
an
article
9 days ago
Article
⚡ nano-vLLM: Lightweight, Low-Latency LLM Inference from Scratch
By
•
•
12
upvoted
an
article
19 days ago
Article
SmolVLM2: Bringing Video Understanding to Every Device
By
and 6 others
•
•
288
upvoted
an
article
about 2 months ago
Article
Introducing Cosmos Predict-2: A Foundation For Your Own World Model
By
and 2 others
•
•
8
upvoted
an
article
2 months ago
Article
🐯 Liger GRPO meets TRL
By
and 5 others
•
•
47
upvoted
an
article
4 months ago
Article
Introduction to 3D Gaussian Splatting
By
•
•
97
upvoted
an
article
6 months ago
Article
Mixture of Experts Explained
By
and 5 others
•
•
788
reacted to
hexgrad's
post with 🔥
6 months ago
Post
21022
📣 Looking for labeled, high-quality synthetic audio/TTS data 📣 Have you been or are you currently calling API endpoints from OpenAI, ElevenLabs, etc? Do you have labeled audio data sitting around gathering dust? Let's talk! Join https://discord.gg/QuGxSWBfQy or comment down below.
If your data exceeds quantity & quality thresholds and is approved into the next hexgrad/Kokoro-82M training mix, and you permissively DM me the data under an effective Apache license, then I will DM back the corresponding voicepacks for YOUR data if/when the next Apache-licensed Kokoro base model drops.
What does this mean? If you've been calling closed-source TTS or audio API endpoints to:
- Build voice agents
- Make long-form audio, like audiobooks or podcasts
- Handle customer support, etc
Then YOU can contribute to the training mix and get useful artifacts in return. ❤️
More details at hexgrad/Kokoro-82M#21
If your data exceeds quantity & quality thresholds and is approved into the next hexgrad/Kokoro-82M training mix, and you permissively DM me the data under an effective Apache license, then I will DM back the corresponding voicepacks for YOUR data if/when the next Apache-licensed Kokoro base model drops.
What does this mean? If you've been calling closed-source TTS or audio API endpoints to:
- Build voice agents
- Make long-form audio, like audiobooks or podcasts
- Handle customer support, etc
Then YOU can contribute to the training mix and get useful artifacts in return. ❤️
More details at hexgrad/Kokoro-82M#21
- 23 replies
upvoted
an
article
11 months ago
Article
What is going on with AlphaFold3?
By
•
•
15