Loading video...

Video Failed to Load

Go Home

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive...

akshay_pachaar's profile picture

Akshay 🚀

279,649 subscribers

126,077 views • 7 months ago •via X (Twitter)

Science & Technology
LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

A peanut-sized Chinese model just dethroned Gemini at reading documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.
0:15

A peanut-sized Chinese model just dethroned Gemini at reading documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.

AlphaSignal AI

91,821 views • 2 months ago

A peanut-sized Chinese model just dethroned Gemini at reading documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.
0:15

A peanut-sized Chinese model just dethroned Gemini at reading documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.

Jafar Najafov

13,630 views • 2 months ago

This peanut-sized chinese model just dethroned Gemini at reading documents. It’s called glm-ocr. it’s a tiny 0.9b parameter vision-language model that is about to replace every expensive ocr api you use. → Handles text, tables, formulas, handwriting → Scored 94.62 on OmniDocBench V1.5 → 8 languages → vLLM, SGLang, Ollama And it’s 100% open-source.
0:15

This peanut-sized chinese model just dethroned Gemini at reading documents. It’s called glm-ocr. it’s a tiny 0.9b parameter vision-language model that is about to replace every expensive ocr api you use. → Handles text, tables, formulas, handwriting → Scored 94.62 on OmniDocBench V1.5 → 8 languages → vLLM, SGLang, Ollama And it’s 100% open-source.

Alvaro Cintas

37,499 views • 2 months ago

GutenOCR: A Grounded Vision-Language Front-End for Documents Fine-tunes Qwen2.5-VL-3B and 7B using a grounded OCR evaluation protocol, substantially improving region- and line-level OCR as well as text-detection recall
0:15

GutenOCR: A Grounded Vision-Language Front-End for Documents Fine-tunes Qwen2.5-VL-3B and 7B using a grounded OCR evaluation protocol, substantially improving region- and line-level OCR as well as text-detection recall

Niels Rogge

45,811 views • 5 months ago

The leading model runnable on a single cloud H100 GPU now fits on a single home GPU! 🔥 We've optimized Gemma 3 27B with QAT so you can run our best-in-class open model on your desktop RTX 3090 or similar. See how easy it is to try via ollama! 👇
0:16

The leading model runnable on a single cloud H100 GPU now fits on a single home GPU! 🔥 We've optimized Gemma 3 27B with QAT so you can run our best-in-class open model on your desktop RTX 3090 or similar. See how easy it is to try via ollama! 👇

Glenn Cameron Jr

81,427 views • 1 year ago

Run Gemma 4 26B MoE on 8GB VRAM with 250k context at 20+ tokens/sec If you own any 8GB VRAM graphics card, stop what you are doing. Local AI just had its absolute "Holy Shit" moment for budget hardware. Yesterday, I benchmarked Unsloth Gemma 4 12B Q4_K_XL on an 8GB card. The community went wild but immediately demanded more: "Can we run a 25B+ model on budget GPUs?" Today, I’m delivering exactly that. I am running a massive 26B parameter Mixture of Experts (MoE) model locally on a standard 8GB VRAM setup with 250k full native context!. If you own an RTX 3060, 3070, 4060, or any budget GPU with 8GB of VRAM, the local AI paradigm has completely changed. The performance metrics are astonishing: - 20 tokens/sec flat decode throughput. - Stable, flat decode speed even with massive prompts. - I threw a 60k token prompt at it, and it still clocked in at 20 TPS without dropping a single frame. # What about prefill? Yes, Time To First Token (TTFT) is slightly high when swallowing massive contexts. But with a solid 200 tokens/sec prefill speed, the wait is barely noticeable and highly usable. And this is running completely without Multi Token Prediction (MTP) active. How is this possible? It’s the magic of Google's new QAT (Quantization Aware Training) quants for Gemma 4. The model weight file (unsloth gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf) is only 13.2 GB, making it the ultimate local powerhouse. # The Test Setup: CPU: Intel Core i7 RAM: 16GB System RAM GPU: NVIDIA GeForce RTX 4060 Laptop GPU (8GB VRAM) # The Secret Sauce (The -cmoe Flag) To make this work properly on any 8GB card, you must use the -cmoe (CPU MoE) flag in llama.cpp. This flag isolates the heavy MoE expert weights directly to system memory (CPU/RAM) while letting your GPU focus strictly on the Attention layers and the KV Cache. It prevents VRAM spillage and holds the throughput rock solid. # The flags: -m "gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf" -cmoe -c 248000 -v Once running, just open the UI on localhost and toggle the new reasoning lightbulb icon in the text input box to watch the model perform multi step thinking. Are you still running smaller models, or are you ready to scale up your budget local setups? Let's discuss in the replies
0:19

Run Gemma 4 26B MoE on 8GB VRAM with 250k context at 20+ tokens/sec If you own any 8GB VRAM graphics card, stop what you are doing. Local AI just had its absolute "Holy Shit" moment for budget hardware. Yesterday, I benchmarked Unsloth Gemma 4 12B Q4_K_XL on an 8GB card. The community went wild but immediately demanded more: "Can we run a 25B+ model on budget GPUs?" Today, I’m delivering exactly that. I am running a massive 26B parameter Mixture of Experts (MoE) model locally on a standard 8GB VRAM setup with 250k full native context!. If you own an RTX 3060, 3070, 4060, or any budget GPU with 8GB of VRAM, the local AI paradigm has completely changed. The performance metrics are astonishing: - 20 tokens/sec flat decode throughput. - Stable, flat decode speed even with massive prompts. - I threw a 60k token prompt at it, and it still clocked in at 20 TPS without dropping a single frame. # What about prefill? Yes, Time To First Token (TTFT) is slightly high when swallowing massive contexts. But with a solid 200 tokens/sec prefill speed, the wait is barely noticeable and highly usable. And this is running completely without Multi Token Prediction (MTP) active. How is this possible? It’s the magic of Google's new QAT (Quantization Aware Training) quants for Gemma 4. The model weight file (unsloth gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf) is only 13.2 GB, making it the ultimate local powerhouse. # The Test Setup: CPU: Intel Core i7 RAM: 16GB System RAM GPU: NVIDIA GeForce RTX 4060 Laptop GPU (8GB VRAM) # The Secret Sauce (The -cmoe Flag) To make this work properly on any 8GB card, you must use the -cmoe (CPU MoE) flag in llama.cpp. This flag isolates the heavy MoE expert weights directly to system memory (CPU/RAM) while letting your GPU focus strictly on the Attention layers and the KV Cache. It prevents VRAM spillage and holds the throughput rock solid. # The flags: -m "gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf" -cmoe -c 248000 -v Once running, just open the UI on localhost and toggle the new reasoning lightbulb icon in the text input box to watch the model perform multi step thinking. Are you still running smaller models, or are you ready to scale up your budget local setups? Let's discuss in the replies

Alok

291,095 views • 26 days ago

NVIDIA just made AI detect objects 10x faster by deleting one step. It's called LocateAnything, and it removes the biggest bottleneck no one else was fixing in vision-language models. Normally a model builds each bounding box one coordinate token at a time. 100 objects means thousands of tokens before an answer. NVIDIA scrapped that: their Parallel Box Decoding predicts the whole box in a single forward pass, as one atomic unit. → 12.7 boxes/sec on one H100 → 10x faster than Qwen3-VL → +3.8% F1 on LVIS, accuracy up, not down → 3B params, runs on one consumer GPU Treating the box as one unit keeps its coordinates tied together, which is why accuracy climbed instead of falling. One model handles detection, GUI grounding, OCR, and document understanding, ready for computer-use agents, robotics, and document pipelines. 100% open source, weights, code, demo, and paper all live.
0:16

NVIDIA just made AI detect objects 10x faster by deleting one step. It's called LocateAnything, and it removes the biggest bottleneck no one else was fixing in vision-language models. Normally a model builds each bounding box one coordinate token at a time. 100 objects means thousands of tokens before an answer. NVIDIA scrapped that: their Parallel Box Decoding predicts the whole box in a single forward pass, as one atomic unit. → 12.7 boxes/sec on one H100 → 10x faster than Qwen3-VL → +3.8% F1 on LVIS, accuracy up, not down → 3B params, runs on one consumer GPU Treating the box as one unit keeps its coordinates tied together, which is why accuracy climbed instead of falling. One model handles detection, GUI grounding, OCR, and document understanding, ready for computer-use agents, robotics, and document pipelines. 100% open source, weights, code, demo, and paper all live.

Alvaro Cintas

198,417 views • 6 days ago

Check this!! A 100% open-source toolkit to work with LLMs. Transformer Lab is an app to experiment with LLMs: - Train, fine-tune, or chat. - One-click LLM download (DeepSeek, Gemma, etc.) - Drag-n-drop UI for RAG. - Built-in logging, and more. 100% local!
0:17

Check this!! A 100% open-source toolkit to work with LLMs. Transformer Lab is an app to experiment with LLMs: - Train, fine-tune, or chat. - One-click LLM download (DeepSeek, Gemma, etc.) - Drag-n-drop UI for RAG. - Built-in logging, and more. 100% local!

Avi Chawla

74,778 views • 1 year ago

Today Meta released "Code Llama", a large language model fine-tuned for coding tasks. It's publicly available and can be used for commercial use! It outperforms GPT 3.5 and you can even run it locally on your Macbook using Ollama.
0:11

Today Meta released "Code Llama", a large language model fine-tuned for coding tasks. It's publicly available and can be used for commercial use! It outperforms GPT 3.5 and you can even run it locally on your Macbook using Ollama.

Marcel Pociot 🧪

50,094 views • 2 years ago

I just created my own LaTeX-OCR app using Llama 3.2 Vision! Upload the LaTeX code as an image, and it gives you the corresponding LaTeX code using Llama 3.2 multimodal! Here's what I used: - Ollama for serving Llama 3.2 vision locally - Streamlit for the UI Everything is just 50 lines of code! Find the code in the next tweet. -- Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs. Find me → Avi Chawla
0:15

I just created my own LaTeX-OCR app using Llama 3.2 Vision! Upload the LaTeX code as an image, and it gives you the corresponding LaTeX code using Llama 3.2 multimodal! Here's what I used: - Ollama for serving Llama 3.2 vision locally - Streamlit for the UI Everything is just 50 lines of code! Find the code in the next tweet. -- Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs. Find me → Avi Chawla

Avi Chawla

15,610 views • 1 year ago

Announcing llama-ocr – a free + open source OCR tool! It takes documents (images for now) & outputs markdown, and does really well for complex receipts, PDFs with tables/charts, ect... Powered by Llama 3.2 vision on Together AI & available on npm today!
0:13

Announcing llama-ocr – a free + open source OCR tool! It takes documents (images for now) & outputs markdown, and does really well for complex receipts, PDFs with tables/charts, ect... Powered by Llama 3.2 vision on Together AI & available on npm today!

Hassan

443,826 views • 1 year ago

Advent of Claude Day 1 - The ! Prefix Don't waste tokens asking "can you run git status or tests?" Just type: ! followed by your bash command The ! prefix executes bash instantly and injects the output into context. No model processing. No delay. No wasted tokens.
0:16

Advent of Claude Day 1 - The ! Prefix Don't waste tokens asking "can you run git status or tests?" Just type: ! followed by your bash command The ! prefix executes bash instantly and injects the output into context. No model processing. No delay. No wasted tokens.

Ado

106,397 views • 7 months ago

📈 Copilot in Word gives you more control to fine-tune your document. Easily edit text, lists, and tables like a pro. It's like an assistant making your job easier. Available on the web and desktop: #AI
0:15

📈 Copilot in Word gives you more control to fine-tune your document. Easily edit text, lists, and tables like a pro. It's like an assistant making your job easier. Available on the web and desktop: #AI

Microsoft 365

19,264 views • 1 year ago

DeepSeek R1 is *the* best model available right now. It's at the level of o1, but you can use it for free, and it's much faster. A huge leap forward that nobody saw coming. No wonder so many people are throwing tantrums online trying to discredit the Chinese students who built this. You can use DeepSeek in Visual Studio Code right now: 1. Install the Qodo Gen AI extension 2. Select DeepSeek R1 from their list of models The Qodo team is hosting DeepSeek on their servers, so none of your data will go to China. I've been building a Tetris game using DeepSeek, and this is the most impressive model I've seen so far.
0:17

DeepSeek R1 is *the* best model available right now. It's at the level of o1, but you can use it for free, and it's much faster. A huge leap forward that nobody saw coming. No wonder so many people are throwing tantrums online trying to discredit the Chinese students who built this. You can use DeepSeek in Visual Studio Code right now: 1. Install the Qodo Gen AI extension 2. Select DeepSeek R1 from their list of models The Qodo team is hosting DeepSeek on their servers, so none of your data will go to China. I've been building a Tetris game using DeepSeek, and this is the most impressive model I've seen so far.

Santiago

1,223,978 views • 1 year ago

PEOPLE ARE BURNING THROUGH THEIR TOKENS IN 30 MINUTES AND BLAMING CLAUDE The problem isn’t with the model. The problem is how they’re using it A simple formatting question they set Max effort The 40th message in the chat the model re-reads the entire history from the beginning Every single time. Using your tokens Low effort costs 10 times less than Max Just get into the habit of routing tasks by level and your limit won’t run out by noon Tokens don’t disappear. You’re just paying for unnecessary processing
0:18

PEOPLE ARE BURNING THROUGH THEIR TOKENS IN 30 MINUTES AND BLAMING CLAUDE The problem isn’t with the model. The problem is how they’re using it A simple formatting question they set Max effort The 40th message in the chat the model re-reads the entire history from the beginning Every single time. Using your tokens Low effort costs 10 times less than Max Just get into the habit of routing tasks by level and your limit won’t run out by noon Tokens don’t disappear. You’re just paying for unnecessary processing

Asteri

15,752 views • 1 month ago

We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025! Paper: Code: Biological systems are capable of rapid adaptation, given limited sensory cues. For example, our human visual system can quickly adapt and tune its light sensitivity to our surroundings. While modern LLMs exhibit a wide variety of capabilities and knowledge, they remain rigid when adding task-specific capabilities. Traditionally, customizing these models requires gathering large datasets and performing often expensive, time-consuming fine-tuning for specific applications. To bypass these limitations, Text-to-LoRA (T2L) meta-learns a “hypernetwork” that takes in a text description of a desired task, as a prompt, and generates a task-specific LoRA that performs well on the task. In our experiments, we show that T2L can encode hundreds of existing LoRA adapters. While the compression is lossy, T2L maintains the performance of task-specifically tuned LoRA adapters. We also show that T2L can even generalize to unseen tasks given a natural language description of the tasks. Importantly, Text-to-LoRA is parameter-efficient. It generates LoRAs in a single, inexpensive step, based solely on a simple text description of the task. This approach is a step towards dramatically lowering the technical and computational barriers, allowing non-technical users to specialize foundation models using plain language, rather than needing deep technical expertise or large compute resources.
0:18

We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025! Paper: Code: Biological systems are capable of rapid adaptation, given limited sensory cues. For example, our human visual system can quickly adapt and tune its light sensitivity to our surroundings. While modern LLMs exhibit a wide variety of capabilities and knowledge, they remain rigid when adding task-specific capabilities. Traditionally, customizing these models requires gathering large datasets and performing often expensive, time-consuming fine-tuning for specific applications. To bypass these limitations, Text-to-LoRA (T2L) meta-learns a “hypernetwork” that takes in a text description of a desired task, as a prompt, and generates a task-specific LoRA that performs well on the task. In our experiments, we show that T2L can encode hundreds of existing LoRA adapters. While the compression is lossy, T2L maintains the performance of task-specifically tuned LoRA adapters. We also show that T2L can even generalize to unseen tasks given a natural language description of the tasks. Importantly, Text-to-LoRA is parameter-efficient. It generates LoRAs in a single, inexpensive step, based solely on a simple text description of the task. This approach is a step towards dramatically lowering the technical and computational barriers, allowing non-technical users to specialize foundation models using plain language, rather than needing deep technical expertise or large compute resources.

Sakana AI

403,067 views • 1 year ago

Gemma 4 is here! Our most intelligent open models to date, are built on the same world-class research and tech as Gemini 3, and are sized to run and fine-tune efficiently on local hardware. Check out what Google Gemma 4 brings to devs: 💎 Advanced Reasoning: Deep logic tasks, complex multi-step planning, and beyond 💎 Longer context: Seamlessly analyze entire codebases with context windows of 128K tokens for our edge models and 256K tokens for our largest models 💎 Vision and audio: Rich, multimodal interactions out of the box 💎 140+ languages: Trained on 140+ languages 💎 Apache 2.0 license: industry-standard open-source license
0:19

Gemma 4 is here! Our most intelligent open models to date, are built on the same world-class research and tech as Gemini 3, and are sized to run and fine-tune efficiently on local hardware. Check out what Google Gemma 4 brings to devs: 💎 Advanced Reasoning: Deep logic tasks, complex multi-step planning, and beyond 💎 Longer context: Seamlessly analyze entire codebases with context windows of 128K tokens for our edge models and 256K tokens for our largest models 💎 Vision and audio: Rich, multimodal interactions out of the box 💎 140+ languages: Trained on 140+ languages 💎 Apache 2.0 license: industry-standard open-source license

Google for Developers

269,578 views • 3 months ago

Build your own DeepSeek based Perplexity!! We built an AI Research Agent that can analyse and research topics, and then it creates a detailed report on Google Docs - Deepseek analyses the topic - Generates a list of questions - Directs another model to research answers to these questions - Report is generated on based on this research Built with Composio LlamaIndex 🦙 Groq Inc Exa Here's the code:
0:16

Build your own DeepSeek based Perplexity!! We built an AI Research Agent that can analyse and research topics, and then it creates a detailed report on Google Docs - Deepseek analyses the topic - Generates a list of questions - Directs another model to research answers to these questions - Report is generated on based on this research Built with Composio LlamaIndex 🦙 Groq Inc Exa Here's the code:

Karan Vaidya

39,566 views • 1 year ago

this is the BEST vision language model I have ever tried! Aria is a new model by Rhymes.AI: a 25.3B multimodal model that can take image/video inputs 🤩 They release the model with Apache-2.0 license and fine-tuning scripts as well 👏 I tested it extensively, keep reading to learn more 🧶
0:10

this is the BEST vision language model I have ever tried! Aria is a new model by Rhymes.AI: a 25.3B multimodal model that can take image/video inputs 🤩 They release the model with Apache-2.0 license and fine-tuning scripts as well 👏 I tested it extensively, keep reading to learn more 🧶

merve

176,047 views • 1 year ago

You can now connect your OpenRouter URL directly to the Warp Agent. OpenRouter unlocks a world of models including Anthropic, OpenAI, Gemini, DeepSeek, Kimi, and more, all with better prices and uptime. Set your URL, model, and an alias to easily switch using /model.
0:17

You can now connect your OpenRouter URL directly to the Warp Agent. OpenRouter unlocks a world of models including Anthropic, OpenAI, Gemini, DeepSeek, Kimi, and more, all with better prices and uptime. Set your URL, model, and an alias to easily switch using /model.

Warp

15,019 views • 1 month ago