[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] #  @MaziyarPanahi Maziyar PANAHI Maziyar PANAHI posts on X about all the, has been, money, failed to the most. They currently have XXXXX followers and XXX posts still getting attention that total XXXXX engagements in the last XX hours. ### Engagements: XXXXX [#](/creator/twitter::206195705/interactions)  - X Week XXXXXXX -XX% - X Month XXXXXXXXX +550% - X Months XXXXXXXXX +99% - X Year XXXXXXXXX +315% ### Mentions: XX [#](/creator/twitter::206195705/posts_active)  - X Week XXX +3.90% - X Month XXX +257% - X Months XXX +119% - X Year XXXXX +158% ### Followers: XXXXX [#](/creator/twitter::206195705/followers)  - X Week XXXXX +3.50% - X Month XXXXX +9% - X Months XXXXX +59% - X Year XXXXX +133% ### CreatorRank: XXXXXXX [#](/creator/twitter::206195705/influencer_rank)  ### Social Influence [#](/creator/twitter::206195705/influence) --- **Social category influence** [technology brands](/list/technology-brands) XXXX% [finance](/list/finance) XXXX% [stocks](/list/stocks) XXXX% [travel destinations](/list/travel-destinations) XXXX% [celebrities](/list/celebrities) XXXX% [cryptocurrencies](/list/cryptocurrencies) XXXX% **Social topic influence** [all the](/topic/all-the) 1.52%, [has been](/topic/has-been) 1.52%, [money](/topic/money) 1.52%, [failed to](/topic/failed-to) 1.52%, [compact](/topic/compact) 0.76%, [gpu](/topic/gpu) 0.76%, [gpt5](/topic/gpt5) 0.76%, [msft](/topic/msft) 0.76%, [msi](/topic/msi) 0.76%, [loops](/topic/loops) XXXX% **Top accounts mentioned or mentioned by** [@huggingface](/creator/undefined) [@primeintellect](/creator/undefined) [@chrisbolas](/creator/undefined) [@vllmproject](/creator/undefined) [@kuittinenpetri](/creator/undefined) [@xainovator](/creator/undefined) [@alibabaqwen](/creator/undefined) [@kpmeister](/creator/undefined) [@nanonets](/creator/undefined) [@stablequan](/creator/undefined) [@oddworldabes](/creator/undefined) [@aonsayyed](/creator/undefined) [@davitausberlin](/creator/undefined) [@kp_meister](/creator/undefined) [@mohammad_53c1](/creator/undefined) [@openmedai](/creator/undefined) [@matvl77](/creator/undefined) [@gheorgheiuga](/creator/undefined) [@anyconvo](/creator/undefined) [@paulmarin90](/creator/undefined) ### Top Social Posts [#](/creator/twitter::206195705/posts) --- Top posts by engagements in the last XX hours "Qwen3-VL-2B the smallest Vision model by @Alibaba_Qwen performing great on doctor's hand writing for its size DAMN ๐ฅ" [X Link](https://x.com/MaziyarPanahi/status/1980686467707072572) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-21T17:23Z 9471 followers, 56.4K engagements "what happened this week with OCR and VLMs * deepseek-ocr * chandra-ocr * nanonets-ocr2 * paddleocr-vl * qwen3-vl (2B 32B Instruct and Thinking) * dots.ocr * olmOCR X (based on Qwen2.5-VL) * LightOnOCR (smallies) top X trending models on @huggingface are still OCR/VLM" [X Link](https://x.com/MaziyarPanahi/status/1981421331053760775) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-23T18:03Z 9472 followers, 65.2K engagements "@alek_k13 i am downloading it now to test" [X Link](https://x.com/MaziyarPanahi/status/1982789844137902365) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T12:41Z 9471 followers, XX engagements "@PrimeIntellect awesome looking forward to all the new goodies please share whenever a bounty is done โ so we can jump and use it ๐" [X Link](https://x.com/MaziyarPanahi/status/1983073380686540951) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T07:28Z 9471 followers, XXX engagements "@casper_hansen_ so msft was holding them back msft has been releasing so many open models themselves. i am not sure what that sentence means to be honest" [X Link](https://x.com/MaziyarPanahi/status/1983235808795496733) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T18:13Z 9472 followers, XXX engagements "i mean this is good for yall at home i play in a different league ๐ 8x H100 2TB memory XX TB enterprise ssd storage 4x 2200W power supply" [X Link](https://x.com/MaziyarPanahi/status/1979903795950657998) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-19T13:33Z 9472 followers, 134.5K engagements "are you shitting me so much for making my own ocr app ๐คฃ i am going to bed. bye. in this photo: me @willccbb @Prince_Canuma @philipvollet" [X Link](https://x.com/MaziyarPanahi/status/1981872081450279200) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-24T23:55Z 9471 followers, 16.4K engagements "@chrisbolas i was just wondering if they are better in any way compared to DGX Spark. but it seems it's more of a branding for those vendors making and selling their own equivalent" [X Link](https://x.com/MaziyarPanahi/status/1982445121502142973) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T13:52Z 9471 followers, XXX engagements "@alamin_ai_ @miramurati exactly i was even thinking distillation at the logits level but it was too expensive. this looks doable ill see if i can implement it this week" [X Link](https://x.com/MaziyarPanahi/status/1982883206795014221) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T18:52Z 9471 followers, XX engagements "๐ Big news in healthcare AI I'm thrilled to announce the launch of OpenMed on @huggingface releasing 380+ state-of-the-art medical NER models for free under Apache XXX. And this is just the beginning ๐งต" [X Link](https://x.com/MaziyarPanahi/status/1945591450084774193) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-07-16T21:08Z 9471 followers, 467.6K engagements "the top two trending models on @huggingface are both for OCR document processing is a hot topic kids ๐" [X Link](https://x.com/MaziyarPanahi/status/1979144220263751705) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-17T11:15Z 9471 followers, 92.5K engagements "RTX Pro 6000 96GB @PrimeIntellect Nanonets-OCR @nanonets vLLM @vllm_project 2000 images from cifar10 @huggingface one word damn" [X Link](https://x.com/MaziyarPanahi/status/1979589541137920132) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-18T16:45Z 9471 followers, 227.4K engagements "i didn't expect "RTX Pro 6000 96GB blackwell" to be this good enough memory fast inference cheap as hell just lovely" [X Link](https://x.com/MaziyarPanahi/status/1979592983096742245) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-18T16:58Z 9471 followers, 286.2K engagements "DeepSeek-OCR on doctor's hand written note" [X Link](https://x.com/MaziyarPanahi/status/1980273244235067892) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-20T14:01Z 9471 followers, 245.6K engagements "Qwen3 VL 4B doctor's hand written note" [X Link](https://x.com/MaziyarPanahi/status/1980404976204808461) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-20T22:45Z 9471 followers, 55.3K engagements "wow deepseek trader just had a huge jump an hour ago ๐ฅ chinese know how to invest" [X Link](https://x.com/MaziyarPanahi/status/1980668748660093205) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-21T16:13Z 9471 followers, 114.2K engagements "isn't this beautiful the new Qwen3 VL 32B explaining grocery store aisle would love to have this on me when i am going shopping ๐ฅ" [X Link](https://x.com/MaziyarPanahi/status/1981007757495894155) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-22T14:40Z 9471 followers, XXX engagements "i am building a simple app (very simple) is HTMX/Alpine.js a good choice for frontend just needs simplicity. plz do not recommend anything with typescript in it. though i love type safe languages i do believe it has no place in js eco system" [X Link](https://x.com/MaziyarPanahi/status/1981343352500211906) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-23T12:54Z 9471 followers, 9678 engagements "whats MSI EdgeXpert is there any relation to DGX Spark" [X Link](https://x.com/MaziyarPanahi/status/1982348160312943082) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T07:26Z 9471 followers, 1708 engagements "wow MiniMax-M2 just dropped 230B MoE with 10B active built for coding agents and tool use MIT license damn. ๐ฅ" [X Link](https://x.com/MaziyarPanahi/status/1982721905326223793) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T08:11Z 9471 followers, 14.6K engagements "@Jafar874 @lifeafterAi_ would be very expensive that much vram in nvidia compared to mac true" [X Link](https://x.com/MaziyarPanahi/status/1982740328529707424) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T09:25Z 9471 followers, XX engagements "@vllm_project you mentioned in the docs we need 4x H200 which makes sense but you also mentioned 4x A100 hows that possible did you mean 8x A100 ps i dont see any quant in the flags so its bf16" [X Link](https://x.com/MaziyarPanahi/status/1982754331574333676) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T10:20Z 9471 followers, 1450 engagements "MiniMaxAI/MiniMax-M2 is giving me roughly XX tokens/s with 4x A100s" [X Link](https://x.com/MaziyarPanahi/status/1982802528224886964) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T13:32Z 9472 followers, 9348 engagements "Most open-source models cant tell if pneumonia in an EHR means the patient has it or not. Thats called Assertion Status the invisible layer that makes clinical AI trustworthy. I'm changing that with open biomedical models on @OpenMed_AI" [X Link](https://x.com/MaziyarPanahi/status/1982908872450003310) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T20:34Z 9471 followers, 1539 engagements "@Shekswess @PrimeIntellect i am planning to use prime-rl today i was using a hackable training loop as kids say it inside verifiers. so i will let you know. but i have always had minimum of X to X gpus for GRPO. never used both inference and training on the same gpu. but it should be doable" [X Link](https://x.com/MaziyarPanahi/status/1983087454719541638) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T08:24Z 9471 followers, XX engagements "just upgraded to macos Tahoe surprised to see you can set up chatgpt in the new xcode XX pretty cool" [X Link](https://x.com/MaziyarPanahi/status/1983195173388734848) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T15:32Z 9471 followers, XXX engagements "@stablequan safe travels my friend" [X Link](https://x.com/MaziyarPanahi/status/1983245361108951142) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T18:51Z 9471 followers, XX engagements "@stablequan looking forward to it enjoy Amsterdam ๐" [X Link](https://x.com/MaziyarPanahi/status/1983247507938271266) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T19:00Z 9471 followers, XX engagements "@BehnamEbrahimi great to hear that whats your primary inputs any multilingual" [X Link](https://x.com/MaziyarPanahi/status/1979436871663952192) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-18T06:38Z 9473 followers, XX engagements "serious talk you can fit a 32B vlm/ocr model easily on one RTX Pro 6000 it costs $X an hour you can easily parse more pages than you can via closed apis it will be also cheaper so are accuracy and ease of use the only reasons you wouldnt host your own ocr model" [X Link](https://x.com/MaziyarPanahi/status/1979595249455612016) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-18T17:07Z 9459 followers, 41.3K engagements "can we use RTX Pro 6000 96GB to play pc games Cyberpunk 2077 ๐" [X Link](https://x.com/MaziyarPanahi/status/1979841840577905030) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-19T09:27Z 9453 followers, 3810 engagements "@matvl77 but 3xVram is a big deal no you have to otherwise buy 3x 5090 to get that much vram" [X Link](https://x.com/MaziyarPanahi/status/1979849053824487489) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-19T09:56Z 9453 followers, 1021 engagements "@ai_for_success the bottom X - one does anything for money - one does anything a political party tells him to do - and one is so delusional about his ai safety he goes around and partners with palantir" [X Link](https://x.com/MaziyarPanahi/status/1979932688682225867) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-19T15:28Z 9471 followers, 32.2K engagements "@lordsigma i am now so excited please do share and take photos its not everyday when you see a storage that size for a single project my largest storage is on 370TB at work a mix of everything and 450TB in our private OpenStack cloud ๐" [X Link](https://x.com/MaziyarPanahi/status/1979999596559335469) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-19T19:54Z 9471 followers, XXX engagements "๐จ RED ALRET ๐จ i don't want to have fun alone so fuck it i am launching a RTX Pro 6000 and giving everyone access here's all you need to use the new Qwen3-VL-2B model free ๐ฅ link to Colab for you ๐" [X Link](https://x.com/MaziyarPanahi/status/1980692255414628637) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-21T17:46Z 9431 followers, 7763 engagements "@Emmilatan awesome let us know how it goes ๐" [X Link](https://x.com/MaziyarPanahi/status/1980975186238173625) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-22T12:31Z 9432 followers, XX engagements "listen to that H100 node powering up" [X Link](https://x.com/MaziyarPanahi/status/1981315655451206045) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-23T11:03Z 9463 followers, 6432 engagements "the good recommendations so far: - pure html+tailwindcss and good old js - FastHTML - just alpine.js" [X Link](https://x.com/MaziyarPanahi/status/1981478986178859449) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-23T21:52Z 9454 followers, XXX engagements "meet Gemini Flash it will be my baseline. im gonna test all the new OCR and VLM models released this week. theyd better handle tool calling well because Im about to summon the BOUNDING_BOX_TOOL to detect objects" [X Link](https://x.com/MaziyarPanahi/status/1981625131169407038) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-24T07:33Z 9467 followers, 7424 engagements "so who's team and who's team ๐" [X Link](https://x.com/MaziyarPanahi/status/1982017366310490362) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-25T09:32Z 9430 followers, XXX engagements "@olesbsn depends in the model but if its good yes" [X Link](https://x.com/MaziyarPanahi/status/1982025021188083922) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-25T10:02Z 9438 followers, XXX engagements "@GUT_AI_F @DOGEFla @AnthropicAI @davidsacks47 @elonmusk this is not new since office XXX you MUST pay per user in your company that's it. just pay don't say BS here about woke don't be cheap pay what you have to pay" [X Link](https://x.com/MaziyarPanahi/status/1982434948541026436) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T13:11Z 9467 followers, XX engagements "i need help from my Chinese followers how do you rate each OCR result Qwen2.5 VL 72B vs Gemini XXX Pro Chinese herbal medicine prescription (might be hard due to being a multilingual not the hand writing perhaps)" [X Link](https://x.com/MaziyarPanahi/status/1982444856082407828) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T13:51Z 9463 followers, 1568 engagements "@chrisbolas yes we can will see how to use them locally via vllm and do the same tests" [X Link](https://x.com/MaziyarPanahi/status/1982461438334496894) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T14:56Z 9455 followers, XX engagements "that's an insane amount of cache ๐" [X Link](https://x.com/MaziyarPanahi/status/1982515436718903762) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T18:31Z 9472 followers, XXX engagements "@Yuchenj_UW i am all about blockchain a big member of DeSci community that relies on decentralized technologies like blockchain IPS DAO etc" [X Link](https://x.com/MaziyarPanahi/status/1982517910561288551) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T18:41Z 9455 followers, XXX engagements "and DeepSeek-OCR unfortunately performs less than Qwen3 VL on handwriting at least. sorry all i know we were all excited. best in my tests are: - qwen2.5 vl - qwen3 vl - deepseek-ocr qwen team has been realising vision models for a long time the level of experience" [X Link](https://x.com/MaziyarPanahi/status/1982525465928831224) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T19:11Z 9473 followers, 3009 engagements "@StijnSmits chatgpt said archive of dependencies in archive format for later" [X Link](https://x.com/MaziyarPanahi/status/1982540701914304763) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T20:11Z 9472 followers, XX engagements "@sinenomine267 these are cloud instances with gpus. the root is like 90G but with 4TB somewhere else. and everything goes to root by default. a better law should be use only XX% of the entire root cause you cannot be more important than the system itself" [X Link](https://x.com/MaziyarPanahi/status/1982569810300178636) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-26T22:07Z 9446 followers, XX engagements "@dtkhangbkg beautiful we should revive our dm groups with the gang see what we can do together for fun" [X Link](https://x.com/MaziyarPanahi/status/1982737628136087728) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T09:14Z 9453 followers, XX engagements "@J3rryH0well @mudler_it Europe" [X Link](https://x.com/MaziyarPanahi/status/1982751957539521017) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T10:11Z 9446 followers, XX engagements "@KuittinenPetri the most useful feature in the reasoning apart from solving phd level is tool use knowing how to use tools reason over the results and be autonomous in a very long turn thats what i want. o3 level otherwise almost XX% of what i need can be covered via Instruct models" [X Link](https://x.com/MaziyarPanahi/status/1982866544490426379) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T17:46Z 9472 followers, XX engagements "am i the only one who thinks typescript is too much and unnecessary for JavaScript/frontend never liked it. always seemed like a waste of time at least in small projects" [X Link](https://x.com/MaziyarPanahi/status/1981592180075471262) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-24T05:22Z 9472 followers, 1763 engagements "@KP_meister it is but how small a model can be to extract accurately is the question" [X Link](https://x.com/MaziyarPanahi/status/1982714791212007457) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T07:43Z 9470 followers, XX engagements "ultimate agi test" [X Link](https://x.com/MaziyarPanahi/status/1982816967359603087) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T14:29Z 9472 followers, 1151 engagements "the reasoning traces on MiniMax-M2 is so polished its like they are post-summarized/polished version of Claude or GPT-5 but the thing is the managed to make the model super smart without having to spend 4k tokens on a reasoning before the answer starts. so thats pretty exciting" [X Link](https://x.com/MaziyarPanahi/status/1982863900677734465) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-27T17:36Z 9469 followers, XX engagements "@officialKrishD @OpenMed_AI beautiful love your workflow wish we had those accessible. will ping you to test once i have something ready for assertion status detection" [X Link](https://x.com/MaziyarPanahi/status/1983051474776023318) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T06:01Z 9471 followers, XX engagements "but cloud providers make so much money hosting these open weights models a quick look at the AWS and Azure you can see they are serving every single open model. knowing openai i a highly doubt that was holding them back this is just there to say if they want it won't be an issue" [X Link](https://x.com/MaziyarPanahi/status/1983260013406867771) [@MaziyarPanahi](/creator/x/MaziyarPanahi) 2025-10-28T19:50Z 9469 followers, XX engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@MaziyarPanahi Maziyar PANAHIMaziyar PANAHI posts on X about all the, has been, money, failed to the most. They currently have XXXXX followers and XXX posts still getting attention that total XXXXX engagements in the last XX hours.
Social category influence technology brands XXXX% finance XXXX% stocks XXXX% travel destinations XXXX% celebrities XXXX% cryptocurrencies XXXX%
Social topic influence all the 1.52%, has been 1.52%, money 1.52%, failed to 1.52%, compact 0.76%, gpu 0.76%, gpt5 0.76%, msft 0.76%, msi 0.76%, loops XXXX%
Top accounts mentioned or mentioned by @huggingface @primeintellect @chrisbolas @vllmproject @kuittinenpetri @xainovator @alibabaqwen @kpmeister @nanonets @stablequan @oddworldabes @aonsayyed @davitausberlin @kp_meister @mohammad_53c1 @openmedai @matvl77 @gheorgheiuga @anyconvo @paulmarin90
Top posts by engagements in the last XX hours
"Qwen3-VL-2B the smallest Vision model by @Alibaba_Qwen performing great on doctor's hand writing for its size DAMN ๐ฅ"
X Link @MaziyarPanahi 2025-10-21T17:23Z 9471 followers, 56.4K engagements
"what happened this week with OCR and VLMs * deepseek-ocr * chandra-ocr * nanonets-ocr2 * paddleocr-vl * qwen3-vl (2B 32B Instruct and Thinking) * dots.ocr * olmOCR X (based on Qwen2.5-VL) * LightOnOCR (smallies) top X trending models on @huggingface are still OCR/VLM"
X Link @MaziyarPanahi 2025-10-23T18:03Z 9472 followers, 65.2K engagements
"@alek_k13 i am downloading it now to test"
X Link @MaziyarPanahi 2025-10-27T12:41Z 9471 followers, XX engagements
"@PrimeIntellect awesome looking forward to all the new goodies please share whenever a bounty is done โ
so we can jump and use it ๐"
X Link @MaziyarPanahi 2025-10-28T07:28Z 9471 followers, XXX engagements
"@casper_hansen_ so msft was holding them back msft has been releasing so many open models themselves. i am not sure what that sentence means to be honest"
X Link @MaziyarPanahi 2025-10-28T18:13Z 9472 followers, XXX engagements
"i mean this is good for yall at home i play in a different league ๐ 8x H100 2TB memory XX TB enterprise ssd storage 4x 2200W power supply"
X Link @MaziyarPanahi 2025-10-19T13:33Z 9472 followers, 134.5K engagements
"are you shitting me so much for making my own ocr app ๐คฃ i am going to bed. bye. in this photo: me @willccbb @Prince_Canuma @philipvollet"
X Link @MaziyarPanahi 2025-10-24T23:55Z 9471 followers, 16.4K engagements
"@chrisbolas i was just wondering if they are better in any way compared to DGX Spark. but it seems it's more of a branding for those vendors making and selling their own equivalent"
X Link @MaziyarPanahi 2025-10-26T13:52Z 9471 followers, XXX engagements
"@alamin_ai_ @miramurati exactly i was even thinking distillation at the logits level but it was too expensive. this looks doable ill see if i can implement it this week"
X Link @MaziyarPanahi 2025-10-27T18:52Z 9471 followers, XX engagements
"๐ Big news in healthcare AI I'm thrilled to announce the launch of OpenMed on @huggingface releasing 380+ state-of-the-art medical NER models for free under Apache XXX. And this is just the beginning ๐งต"
X Link @MaziyarPanahi 2025-07-16T21:08Z 9471 followers, 467.6K engagements
"the top two trending models on @huggingface are both for OCR document processing is a hot topic kids ๐"
X Link @MaziyarPanahi 2025-10-17T11:15Z 9471 followers, 92.5K engagements
"RTX Pro 6000 96GB @PrimeIntellect Nanonets-OCR @nanonets vLLM @vllm_project 2000 images from cifar10 @huggingface one word damn"
X Link @MaziyarPanahi 2025-10-18T16:45Z 9471 followers, 227.4K engagements
"i didn't expect "RTX Pro 6000 96GB blackwell" to be this good enough memory fast inference cheap as hell just lovely"
X Link @MaziyarPanahi 2025-10-18T16:58Z 9471 followers, 286.2K engagements
"DeepSeek-OCR on doctor's hand written note"
X Link @MaziyarPanahi 2025-10-20T14:01Z 9471 followers, 245.6K engagements
"Qwen3 VL 4B doctor's hand written note"
X Link @MaziyarPanahi 2025-10-20T22:45Z 9471 followers, 55.3K engagements
"wow deepseek trader just had a huge jump an hour ago ๐ฅ chinese know how to invest"
X Link @MaziyarPanahi 2025-10-21T16:13Z 9471 followers, 114.2K engagements
"isn't this beautiful the new Qwen3 VL 32B explaining grocery store aisle would love to have this on me when i am going shopping ๐ฅ"
X Link @MaziyarPanahi 2025-10-22T14:40Z 9471 followers, XXX engagements
"i am building a simple app (very simple) is HTMX/Alpine.js a good choice for frontend just needs simplicity. plz do not recommend anything with typescript in it. though i love type safe languages i do believe it has no place in js eco system"
X Link @MaziyarPanahi 2025-10-23T12:54Z 9471 followers, 9678 engagements
"whats MSI EdgeXpert is there any relation to DGX Spark"
X Link @MaziyarPanahi 2025-10-26T07:26Z 9471 followers, 1708 engagements
"wow MiniMax-M2 just dropped 230B MoE with 10B active built for coding agents and tool use MIT license damn. ๐ฅ"
X Link @MaziyarPanahi 2025-10-27T08:11Z 9471 followers, 14.6K engagements
"@Jafar874 @lifeafterAi_ would be very expensive that much vram in nvidia compared to mac true"
X Link @MaziyarPanahi 2025-10-27T09:25Z 9471 followers, XX engagements
"@vllm_project you mentioned in the docs we need 4x H200 which makes sense but you also mentioned 4x A100 hows that possible did you mean 8x A100 ps i dont see any quant in the flags so its bf16"
X Link @MaziyarPanahi 2025-10-27T10:20Z 9471 followers, 1450 engagements
"MiniMaxAI/MiniMax-M2 is giving me roughly XX tokens/s with 4x A100s"
X Link @MaziyarPanahi 2025-10-27T13:32Z 9472 followers, 9348 engagements
"Most open-source models cant tell if pneumonia in an EHR means the patient has it or not. Thats called Assertion Status the invisible layer that makes clinical AI trustworthy. I'm changing that with open biomedical models on @OpenMed_AI"
X Link @MaziyarPanahi 2025-10-27T20:34Z 9471 followers, 1539 engagements
"@Shekswess @PrimeIntellect i am planning to use prime-rl today i was using a hackable training loop as kids say it inside verifiers. so i will let you know. but i have always had minimum of X to X gpus for GRPO. never used both inference and training on the same gpu. but it should be doable"
X Link @MaziyarPanahi 2025-10-28T08:24Z 9471 followers, XX engagements
"just upgraded to macos Tahoe surprised to see you can set up chatgpt in the new xcode XX pretty cool"
X Link @MaziyarPanahi 2025-10-28T15:32Z 9471 followers, XXX engagements
"@stablequan safe travels my friend"
X Link @MaziyarPanahi 2025-10-28T18:51Z 9471 followers, XX engagements
"@stablequan looking forward to it enjoy Amsterdam ๐"
X Link @MaziyarPanahi 2025-10-28T19:00Z 9471 followers, XX engagements
"@BehnamEbrahimi great to hear that whats your primary inputs any multilingual"
X Link @MaziyarPanahi 2025-10-18T06:38Z 9473 followers, XX engagements
"serious talk you can fit a 32B vlm/ocr model easily on one RTX Pro 6000 it costs $X an hour you can easily parse more pages than you can via closed apis it will be also cheaper so are accuracy and ease of use the only reasons you wouldnt host your own ocr model"
X Link @MaziyarPanahi 2025-10-18T17:07Z 9459 followers, 41.3K engagements
"can we use RTX Pro 6000 96GB to play pc games Cyberpunk 2077 ๐"
X Link @MaziyarPanahi 2025-10-19T09:27Z 9453 followers, 3810 engagements
"@matvl77 but 3xVram is a big deal no you have to otherwise buy 3x 5090 to get that much vram"
X Link @MaziyarPanahi 2025-10-19T09:56Z 9453 followers, 1021 engagements
"@ai_for_success the bottom X - one does anything for money - one does anything a political party tells him to do - and one is so delusional about his ai safety he goes around and partners with palantir"
X Link @MaziyarPanahi 2025-10-19T15:28Z 9471 followers, 32.2K engagements
"@lordsigma i am now so excited please do share and take photos its not everyday when you see a storage that size for a single project my largest storage is on 370TB at work a mix of everything and 450TB in our private OpenStack cloud ๐"
X Link @MaziyarPanahi 2025-10-19T19:54Z 9471 followers, XXX engagements
"๐จ RED ALRET ๐จ i don't want to have fun alone so fuck it i am launching a RTX Pro 6000 and giving everyone access here's all you need to use the new Qwen3-VL-2B model free ๐ฅ link to Colab for you ๐"
X Link @MaziyarPanahi 2025-10-21T17:46Z 9431 followers, 7763 engagements
"@Emmilatan awesome let us know how it goes ๐"
X Link @MaziyarPanahi 2025-10-22T12:31Z 9432 followers, XX engagements
"listen to that H100 node powering up"
X Link @MaziyarPanahi 2025-10-23T11:03Z 9463 followers, 6432 engagements
"the good recommendations so far: - pure html+tailwindcss and good old js - FastHTML - just alpine.js"
X Link @MaziyarPanahi 2025-10-23T21:52Z 9454 followers, XXX engagements
"meet Gemini Flash it will be my baseline. im gonna test all the new OCR and VLM models released this week. theyd better handle tool calling well because Im about to summon the BOUNDING_BOX_TOOL to detect objects"
X Link @MaziyarPanahi 2025-10-24T07:33Z 9467 followers, 7424 engagements
"so who's team and who's team ๐"
X Link @MaziyarPanahi 2025-10-25T09:32Z 9430 followers, XXX engagements
"@olesbsn depends in the model but if its good yes"
X Link @MaziyarPanahi 2025-10-25T10:02Z 9438 followers, XXX engagements
"@GUT_AI_F @DOGEFla @AnthropicAI @davidsacks47 @elonmusk this is not new since office XXX you MUST pay per user in your company that's it. just pay don't say BS here about woke don't be cheap pay what you have to pay"
X Link @MaziyarPanahi 2025-10-26T13:11Z 9467 followers, XX engagements
"i need help from my Chinese followers how do you rate each OCR result Qwen2.5 VL 72B vs Gemini XXX Pro Chinese herbal medicine prescription (might be hard due to being a multilingual not the hand writing perhaps)"
X Link @MaziyarPanahi 2025-10-26T13:51Z 9463 followers, 1568 engagements
"@chrisbolas yes we can will see how to use them locally via vllm and do the same tests"
X Link @MaziyarPanahi 2025-10-26T14:56Z 9455 followers, XX engagements
"that's an insane amount of cache ๐"
X Link @MaziyarPanahi 2025-10-26T18:31Z 9472 followers, XXX engagements
"@Yuchenj_UW i am all about blockchain a big member of DeSci community that relies on decentralized technologies like blockchain IPS DAO etc"
X Link @MaziyarPanahi 2025-10-26T18:41Z 9455 followers, XXX engagements
"and DeepSeek-OCR unfortunately performs less than Qwen3 VL on handwriting at least. sorry all i know we were all excited. best in my tests are: - qwen2.5 vl - qwen3 vl - deepseek-ocr qwen team has been realising vision models for a long time the level of experience"
X Link @MaziyarPanahi 2025-10-26T19:11Z 9473 followers, 3009 engagements
"@StijnSmits chatgpt said archive of dependencies in archive format for later"
X Link @MaziyarPanahi 2025-10-26T20:11Z 9472 followers, XX engagements
"@sinenomine267 these are cloud instances with gpus. the root is like 90G but with 4TB somewhere else. and everything goes to root by default. a better law should be use only XX% of the entire root cause you cannot be more important than the system itself"
X Link @MaziyarPanahi 2025-10-26T22:07Z 9446 followers, XX engagements
"@dtkhangbkg beautiful we should revive our dm groups with the gang see what we can do together for fun"
X Link @MaziyarPanahi 2025-10-27T09:14Z 9453 followers, XX engagements
"@J3rryH0well @mudler_it Europe"
X Link @MaziyarPanahi 2025-10-27T10:11Z 9446 followers, XX engagements
"@KuittinenPetri the most useful feature in the reasoning apart from solving phd level is tool use knowing how to use tools reason over the results and be autonomous in a very long turn thats what i want. o3 level otherwise almost XX% of what i need can be covered via Instruct models"
X Link @MaziyarPanahi 2025-10-27T17:46Z 9472 followers, XX engagements
"am i the only one who thinks typescript is too much and unnecessary for JavaScript/frontend never liked it. always seemed like a waste of time at least in small projects"
X Link @MaziyarPanahi 2025-10-24T05:22Z 9472 followers, 1763 engagements
"@KP_meister it is but how small a model can be to extract accurately is the question"
X Link @MaziyarPanahi 2025-10-27T07:43Z 9470 followers, XX engagements
"ultimate agi test"
X Link @MaziyarPanahi 2025-10-27T14:29Z 9472 followers, 1151 engagements
"the reasoning traces on MiniMax-M2 is so polished its like they are post-summarized/polished version of Claude or GPT-5 but the thing is the managed to make the model super smart without having to spend 4k tokens on a reasoning before the answer starts. so thats pretty exciting"
X Link @MaziyarPanahi 2025-10-27T17:36Z 9469 followers, XX engagements
"@officialKrishD @OpenMed_AI beautiful love your workflow wish we had those accessible. will ping you to test once i have something ready for assertion status detection"
X Link @MaziyarPanahi 2025-10-28T06:01Z 9471 followers, XX engagements
"but cloud providers make so much money hosting these open weights models a quick look at the AWS and Azure you can see they are serving every single open model. knowing openai i a highly doubt that was holding them back this is just there to say if they want it won't be an issue"
X Link @MaziyarPanahi 2025-10-28T19:50Z 9469 followers, XX engagements
/creator/twitter::MaziyarPanahi