Shivay Lamba @HowDevelop

AI Architect & Builder - Inference, LLMs, SLMs GSoC Org Admin - Jenkins | Docker Captain | CNCF Ambassador shivaylamba.me New Delhi Joined December 2018

Tweets

6K
Followers

8K
Following

1K
Likes

34K

h100envy @h100envy

2 days ago

Ex-NVIDIA engineer who built Unsloth explained RL, kernels, reasoning, quantization, and agents in 2 hours 42 minutes - better than $5000 fine-tuning bootcamps. pick the base model -> write triton kernels for 2x faster fine-tune -> quantize to 4-bit -> run GRPO/DPO -> ship a reasoning model on your single GPU. That loop is why Unsloth is the default way to fine-tune Llama, Qwen, Gemma, and Phi on hardware you already own. Unsloth + Triton kernels + 4-bit quantization + GRPO/DPO + single-GPU fine-tuning - that's the stack. Watch and save it, then fine-tune your first model tonight.

h100envy @h100envy

3 weeks ago

x.com/i/article/2068…

12 68 536 812K 1K

42 494 4K 362K 8K

View Details

Shivay Lamba @HowDevelop

7 hours ago

@HaimantikaM Dosa is a cheat meal? :o

1 0 0 82 0

View Details

Shivay Lamba @HowDevelop

19 hours ago

@krisvelkov @AakanshaPriya_ amazing people!@

1 0 3 168 0

View Details

Shivay Lamba @HowDevelop

19 hours ago

@aaishika @Qualcomm_Dev so coooooool

0 0 2 160 0

View Details

clem 🤗 @ClementDelangue

a day ago

Keeping up with AI news is becoming a full-time job. So my friend @ivan_bezdomny built HuggingNews, an AI-curated feed that surfaces the news actually worth reading. Soon, it will even personalize the feed using your Hugging Face profile. Been using it for weeks! Bookmark it, or ask your agent to send you the top 10 stories every morning or night. Less noise. More signal. More building! huggingnews.com

104 201 2K 126K 2K

View Details

Parampreet Singh @Param3021

a day ago

Want to build something with ChatGPT Codex App? - With 5.6 Sol? Register for Codex Hackathon happening on July 14 in New Delhi. 🚀 luma.com/yx3p1zxx Closing soon! And RSVPs are rolling out tomorrow! Build With Codex 🚀🚀🚀

Parampreet Singh @Param3021

a week ago

We're back with Codex Hackathon in Delhi NCR! 8 hours hackathon, only for serious builders, who love to ship with codex! 🚀 📍Masters' Union on Tuesday, July 14! Register: luma.com/yx3p1zxx Let’s build with Codex. 🚀🚀🚀 @HowDevelop @gabrielchua @paw_lean @reach_vb

7 6 28 4K 4

2 2 10 652 3

View Details

Shivay Lamba @HowDevelop

a day ago

@LysandreJik @vllm_project @rudrakshkarpe

0 0 1 132 0

View Details

Gabriel Chua @gabrielchua

2 days ago

OpenAI Build Week is GLOBAL 🌍🌎🌏 A huge thank you to our Codex Ambassadors for bringing the community together. Let's build!!! +++++++++++++++++++++++++++ July 13: Brisbane, Australia 🇦🇺 July 13: Buenos Aires, Argentina 🇦🇷 July 13: Dhaka, Bangladesh 🇧🇩 July 13: Salt Lake City, United States 🇺🇸 July 13: San Salvador, El Salvador 🇸🇻 July 13: Seoul, South Korea 🇰🇷 July 13: Taichung, Taiwan 🇹🇼 July 14: Buenos Aires, Argentina 🇦🇷 July 14: Concepción, Chile 🇨🇱 July 14: Dubai, United Arab Emirates 🇦🇪 July 14: Gurugram, India 🇮🇳 July 14: Heilbronn, Germany 🇩🇪 July 14: Kaohsiung, Taiwan 🇹🇼 July 14: Monterrey, Mexico 🇲🇽 July 15: London, United Kingdom 🇬🇧 July 15: Madrid, Spain 🇪🇸 July 15: Manta, Ecuador 🇪🇨 July 15: Melbourne, Australia 🇦🇺 July 15: Osaka, Japan 🇯🇵 July 15: Sydney, Australia 🇦🇺 July 16: Göppingen, Germany 🇩🇪 July 16: Kuala Lumpur, Malaysia 🇲🇾 July 16: Vienna, Austria 🇦🇹 July 17: Ho Chi Minh City, Vietnam 🇻🇳 July 17: Nairobi, Kenya 🇰🇪 July 17: Paris, France 🇫🇷 July 17: San Francisco, United States 🇺🇸 July 18: Bangkok, Thailand 🇹🇭 July 18: Berlin, Germany 🇩🇪 July 18: Brussels, Belgium 🇧🇪 July 18: Chiang Mai, Thailand 🇹🇭 July 18: Dhaka, Bangladesh 🇧🇩 July 18: Ernakulam, India 🇮🇳 July 18: Fayetteville, United States 🇺🇸 July 18: Göppingen, Germany 🇩🇪 July 18: Hanoi, Vietnam 🇻🇳 July 18: Ho Chi Minh City, Vietnam 🇻🇳 July 18: Hyderabad, India 🇮🇳 July 18: Jakarta, Indonesia 🇮🇩 July 18: Kuala Lumpur, Malaysia 🇲🇾 July 18: Lima, Peru 🇵🇪 July 18: London, United Kingdom 🇬🇧 July 18: Manila, Philippines 🇵🇭 July 18: Melbourne, Australia 🇦🇺 July 18: Paris, France 🇫🇷 July 18: Perth, Australia 🇦🇺 July 18: Pune, India 🇮🇳 July 18: Singapore 🇸🇬 July 18: Spokane, United States 🇺🇸 July 18: Tokyo, Japan 🇯🇵 July 18: Vancouver, Canada 🇨🇦 July 19: Christchurch, New Zealand 🇳🇿 July 19: Singapore 🇸🇬 July 19: Sydney, Australia 🇦🇺 July 19: Taipei, Taiwan 🇹🇼 July 20: Athens, Greece 🇬🇷 July 20: Bentonville, United States 🇺🇸 July 20: Dhaka, Bangladesh 🇧🇩 July 20: Santiago, Chile 🇨🇱 July 21: Miami, United States 🇺🇸 July 21: Portoviejo, Ecuador 🇪🇨 July 21: San Diego, United States 🇺🇸 July 21: San Francisco, United States 🇺🇸 July 21: Santiago, Chile 🇨🇱 July 21: São Paulo, Brazil 🇧🇷 July 21: Tel Aviv, Israel 🇮🇱

31 27 413 735K 28

View Details

Vaibhav (VB) Srivastav @reach_vb

a day ago

Reasoning efforts on GPT-5.6 Sol: Start with the lowest setting that gets the job done, then turn it up only when the task actually needs more thinking. > Light & Low: quick, well-scoped tasks > Medium: planning, analysis, or a few moving parts > High & xhigh: harder work with multiple steps, tradeoffs, research, or careful verification Max and Ultra are a bit different: > Max: one model spends longer working through one hard problem > Ultra: multiple subagents work on different parts of a bigger problem in parallel Oh and, as with any new model, the reasoning levels don’t map directly from GPT-5.5 to GPT-5.6. If you’re used to 5.5, try starting one level lower on a familiar task, then increase it only if the result needs it.

103 44 600 71K 257

View Details

Shivay Lamba @HowDevelop

a day ago

@MrAhmadAwais @elonmusk wow this is insane!!!!

1 0 3 260 0

View Details

Shivay Lamba @HowDevelop

2 days ago

@jason_mayes @Google Good to finally see a true spiritual successor to TensorFlowJS!

1 0 3 211 0

View Details

Shivay Lamba @HowDevelop

2 days ago

@yashgogri1 @DhravyaShah would love to try it out as well Yash! Happy to provide feedback :)

1 0 1 46 0

View Details

Shivay Lamba @HowDevelop

2 days ago

@rudrank did you get early access to Sol?😃

0 0 2 18 0

View Details

Shivay Lamba @HowDevelop

2 days ago

@rudrank great read!

1 0 1 132 0

View Details

Shivay Lamba @HowDevelop

2 days ago

@carrycooldude but now hopefully soon, you will have to become ready for even longer travels, if you know what I mean😃

0 0 0 73 0

View Details

Shivay Lamba @HowDevelop

2 days ago

@sama x.com/HowDevelop/sta… Certainly better than 5.5

Shivay Lamba @HowDevelop

2 days ago

0 0 2 635 0

0 0 0 151 0

View Details

Shivay Lamba @HowDevelop

2 days ago

Blame the new models for this cost in 4 days 😂

3 0 12 791 0

View Details

Shivay Lamba @HowDevelop

2 days ago

Continuing from yesterday's @cursor_ai benchmark here's where gpt-5.6-sol-high lands on tokens per dollar? ~1,971 output tok/$ ---> perfectly in the middle. It beats gpt-5.5 and opus-4.8-thinking-high-fast, but grok-4.5 and opus-thinking-high (~2,850–2,890) edge it out, and composer-2.5-fast (~6,950) is still untouchable on value. Performance wise, pretty similar to Fable for harder and long agent running tasks. P.S. personal run

Shivay Lamba @HowDevelop

3 days ago

I have tried out grok 4.5 quite extensively on @cursor_ai in a given repository where I also tested Fable, Opus 4.8 thinking high, 4.8 thinking-high-fast, GPT 5.5 and Composer 2.5 Fast. It was surprising to see that after multiple test runs,looking at the output tokens per

3 1 13 3K 0

0 0 2 635 0

View Details

Shivay Lamba @HowDevelop

2 days ago

@BhosalePratim Just a small correction, on Github, because some user took up the user name Anthropic, the Anthropic team has to use github.com/anthropics 😆

2 0 9 2K 1

View Details

Vaibhav (VB) Srivastav @reach_vb

2 days ago

Excited to announce OpenAI Build Week! A week full of excitement, joy and dedication to pushing what's possible with GPT 5.6 and Codex We're introducing a Global challenge starting July 13 and over 40+ events across the globe to bring this experience closer to you Kudos to our Codex Ambassadors for putting together such tremendous experiences across the planet from SF to Tokyo in the coming week!! Sign up now and excited to see what you build!