AI Education @FireworksAI_HQ 🔥🎆 NLP + Gen AI Expert / LLM whisperer 🤖 AI Author 📚 Founder @Aikylie (acquired) 📈 Fellow @YCombinator
San Francisco, CA
Joined March 2012
- Tweets562
- Following1.6K
- Followers2K
- Likes975
Sinan Ozdemir retweeted
DeepSeek V4.1 Flash 1017 runs at 1,521+ tok/s on Fireworks Ultrafast.
Fast enough to ship Tetris in 11 seconds, 7x faster than Claude Opus 5.5.
Join the Ultrafast waitlist: bit.ly/3UA0wKq
Sinan Ozdemir retweeted
Check out this month's AI Systems Performance Engineering meetup: "Deep dive into Fireworks AI continuously improving their RL post-training through Rollout, Reward, Update, and Repeat!" by @Prof_OZ @FireworksAI_HQ !! with my co-organizer @anbarth
youtu.be/s1j7-x6uk_0
Sinan Ozdemir retweeted
do you know what you pay for in agentic workloads? cached tokens!
session with 50+ tool calls -> prompt is billed 50 times
all providers give 1/5 cached discount for GLM-5.2
we at @FireworksAI_HQ dropped it to 1/10, matching GPT/Claude
that's -40% typical savings, have fun!
Sinan Ozdemir retweeted
DSpark from @deepseek_ai ingeniously integrates many speculative decoding ideas to achieve 1.5x to 5x higher throughput in a real production system
Let's understand it with 10 ideas, starting from the very basics 🧵
Sinan Ozdemir retweeted
In a joint Fireworks and @Faros_AI evaluation of 211 real engineering tasks, Claude Code + GLM-5.2 beat both Claude Code + Opus 4.8 and Codex + GPT-5.5:
- Judge score: 0.568 vs. 0.521 and 0.466
- Time per task: 321s vs. 775s and 392s
- Cost per task: $0.92 vs. $1.76 and $2.06
Most importantly, Faros tested the models on its own repositories and work, not just public benchmarks.
Model choice should be based on where a model clears your bar, on your work, at a cost that makes sense.
With our partner @fireworksai_hq, we ran an evaluation to find out if open models have gotten strong enough to handle real software engineering work. Our result: Claude Code + GLM-5.2 matched Kimi on quality while running faster and cheaper. More details:
faros.ai/blog/open-models-vs…
Sinan Ozdemir retweeted
GLM-5.2 can now be run locally!🔥
The 2-bit model retains ~82% accuracy after we shrunk it from 1.51TB to 238GB (-84% size).
Run on a 256GB Mac or RAM/VRAM setups.
GLM-5.2 is the strongest open model to date.
Guide: unsloth.ai/docs/models/glm-5…
GGUF: huggingface.co/unsloth/GLM-5…
Introducing GLM-5.2: Frontier Intelligence, Open Weights
- Significant improvements in coding and agentic tasks
- Strong long-horizon capabilities with a 1M context window
- Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong balance between performance and token efficiency
- MIT-licensed open weights
- Same API pricing as GLM-5.1
Tech Blog: z.ai/blog/glm-5.2
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.2
Coding Plan: z.ai/subscribe
Chat: chat.z.ai
Sinan Ozdemir retweeted
BTW I've been using using @FireworksAI_HQ for inference. Other providers might not be as fast.
Sinan Ozdemir retweeted
We’re releasing GPT-4 — a large multimodal model (image & text in, text out) which is a significant advance in both capability and alignment.
Still limited in many ways, but passes many qualification benchmarks like the bar exam & AP Calculus: openai.com/research/gpt-4
Sinan Ozdemir retweeted
New open source Flan-UL2 20B checkpoints :)
- Truly open source 😎 No forms! 🤭 Apache license 🔥
- Best OS model on MMLU/Big-Bench hard 🤩
- Better than Flan-T5 XXL & competitive to Flan-PaLM 62B.
- Size ceiling of Flan family just got higher!
Blog: yitay.net/blog/flan-ul2-20b
Sinan Ozdemir retweeted
Over 1400 people registered for yesterday's "NLP with ChatGPT (and other LLMs)" conference that I hosted in the @OReillyMedia platform. Kudos to speakers @Prof_OZ , Melanie and Shaan for making it a smashing success 🎉
#machinelearning #ai #nlp #chatgpt #LLM #python
Sinan Ozdemir retweeted
Is LLM finetuning worth it?
If you know what you're doing, finetuned models can be 30x smaller❗️without losing performance. This can unlock applications that would otherwise be too expensive or slow.
STaR is a great example of doing it right.
(DIY instructions at the end)
Sinan Ozdemir retweeted
🏷️Deal of the Day🛒
How to improve your #machinelearning pipelines without hours spent on fine-tuning parameters?
The answer is #featureengineering: mng.bz/GOpJ
@Prof_OZ #ML #data
Sinan Ozdemir retweeted
For more prompting guidance have a look at humanloop.com/blog/prompt-en… - we hope the tips and tricks help you on your journey of building effective LLM applications.
Thank you to @prof_oz for the guest post.
Sinan Ozdemir retweeted
Master Large Language Models in 2 weeks by @Prof_Oz
For devs and data scientists:
- Use large language models like GPT-3 and Cohere effectively
- Fine-tune open source language models
- Quantify and mitigate bias in NLP pipelines
Enroll by Sep 19: bit.ly/3AKBx9t
Sinan Ozdemir retweeted
🏷️Deal of the Day🛒
Improve your #machinelearning pipelines without hours spent on fine-tuning parameters!
That’s what you’ll get from #featureengineering: mng.bz/GOpJ
@Prof_OZ #ML #data
Sinan Ozdemir retweeted
📢Now in print!🍾
Looking for the next level in ML performance?
Apply #featureengineering to get better results from your #ML pipelines—faster: mng.bz/GOpJ
@Prof_OZ #machinelearning #data
