hrpnk

7 ms·

61.
by hrpnk 1y ago
They will retire lots of models: GPT-4o, GPT-4.1, GPT-4.5, GPT-4.1-mini, o4-mini, o4-mini-high, o3, o3-pro. https://help.openai.com/en/articles/6825453-chatgpt-release-... "If you open a conversation that use
62.
by hrpnk 1y ago
Thank you! Flash attention gives me a boost to ~66 tok/s indeed.
63.
by hrpnk 1y ago
gpt-oss-20b: 9 threads, 131072 context window, 4 experts - 35-37 tok/s on M2 Max via LM Studio.
64.
by hrpnk 1y ago
With LM Studio you can configure context window freely. Max is 131072 for gpt-oss-20b.
65.
by hrpnk 1y ago
I tried to generate a streamlit dashboard with MACD, RSI, MA(200). 1:0 for qwen3 here. qwen3-coder-30b 4-bit mlx took on the task w/o any hiccups with a fully working dashboard, graphs, and recent data fetched from yfinance. gpt-oss-20
66.
by hrpnk 1y ago
Previously it was all sniper and sneaker bots scanning websites for product availability and attempting purchases continuously to snipe when it comes back online. Now, it's a gazillion of AI crawlers and python crawlers, MCP servers th
67.
by hrpnk 1y ago
last time I checked, context7 depended on an opt-in from library authors expressed through a marker file in the repository, which is negatively affecting adoption and docs coverage.
68.
by hrpnk 1y ago
Looks like a v0.dev competitor. Announcement: https://github.blog/changelog/2025-07-23-github-spark-in-pub... Post from Satya with video: https://x.com/satyanadella/status/1948101877486452897
69.
by hrpnk 1y ago
Has anyone seen OTel being used well for long-running batch/async processes? Wonder how the suggestions stack up to monolith builds for Apps that take about an hour.
70.
by hrpnk 1y ago
Wait for the models to be able to learn to estimate the economic value of each issue taking into account 0-day security issues and falling stock prices. They will quote you accordingly with a marked up price. Would definitely sell well when
71.
by hrpnk 1y ago
This critical look focuses just on the protocol. The fun starts with the actual MCP server implementations... Seems that providing an MCP server is the to be or not to be for all sorts of vendors. All REST APIs get wrapped into an MCP to ma
72.
by hrpnk 2y ago
Available on Groq: https://groq.com/llama-4-now-live-on-groq-build-fast-at-the-... Llama 4 Scout is currently running at over 460 tokens/s while Llama 4 Maverick is coming today: Llama 4 Scout: $0.11 / M input tok
73.
by hrpnk 2y ago
What's missing in these examples are evals and any advice on creating a verification set that can be used to assert that the system continues to work as designed. Models and their prompting patterns change; one cannot just pretend that
74.
Blueskyfeedbot: Post RSS Feeds to Bluesky via GitHub Actions(github.com)
154 points by hrpnk 2y ago|70 comments
75.
by hrpnk 2y ago
What is also interesting (and troubling to see) is all the AI influencers panicing and inventing conspiracy theories downplaying the engineering achievements of the team behind Deepseek. Catching up is always easier than cruising by having
76.
by hrpnk 2y ago
Most of the HF models have a code snippet that you can use in order to run inference on the model. The transformers library will take care of the download as a dependency when you run the code. Typically, a python 3.10-3.11 environment is s
77.
by hrpnk 2y ago
Seems that a bunch of quantized models are already uploaded to ollama: https://ollama.com/library/llama3.3/tags
78.
by hrpnk 2y ago
Reading the Llama community license agreement, section "Redistribution and Use" I expected to find 'Built with Llama'. Is this not required? https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Ins
79.
by hrpnk 2y ago
Marc Rubinstein also has great educational articles on history and inner workings of the finance industry: https://www.netinterest.co/
80.
by hrpnk 2y ago
Have you seen https://github.com/zalando/skipper ? You can implement custom filters in Golang.
81.
by hrpnk 2y ago
From a first look, code generation is done with llama 3.1 405B. Screenshot analysis with llama 3.2 90B. Prompts: https://github.com/Nutlope/napkins/blob/f6c89c76b07b234c7ec6...
82.
by hrpnk 2y ago
Is there a reason you didn't just implement OpenTelemetry (OT) straight away? Curious about the trade offs to opt for a home-grown telemetry inspired by OT instead.
83.
by hrpnk 2y ago
Maybe they will allow a level-playing field and allow bids like in a marketplace?
84.
by hrpnk 2y ago
The model's context is 128k tokens, so you'd have to split the data and analyze in chunks.
85.
by hrpnk 2y ago
They mentioned they do dogfooding. Wonder why it did not work for this update.
86.
by hrpnk 2y ago
> Without a way to coordinate merges cross-repository, we would see more CI failures and broken builds due to semantically incompatible patches landing on the mainline branch. Isn't this rather a symptom of code organization or arch
87.
by hrpnk 2y ago
> Look at all the environment variables in this dockerfile — which ones are necessary to run this service myself? It would also help to sort and de-duplicate the variables in the file. COHERE_API_KEY is listed twice. Adding a comment per
88.
by hrpnk 2y ago
I was interested in how the duplicate detection of issues is working. It uses Cohere API for embeddings (full issue text) [1] and vector-based similarity search [2]. [1] https://github.com/tegonhq/tegon/blob/1
89.
by hrpnk 2y ago
There is also COHERE_API_KEY for the vector storage & search: https://github.com/tegonhq/tegon/blob/158b54af8d6f7cf4195c61...
90.
by hrpnk 2y ago
Poetry user here. What is so special about Conda that makes it particularly good for ML use cases?

More ›