_______ __ _______ | | |.---.-..----.| |--..-----..----. | | |.-----..--.--.--..-----. | || _ || __|| < | -__|| _| | || -__|| | | ||__ --| |___|___||___._||____||__|__||_____||__| |__|____||_____||________||_____| on Gopher (inofficial) URI Visit Hacker News on the Web COMMENT PAGE FOR: DIR Show HN: Arch-Router â 1.5B model for LLM routing by preferences, not benchmarks pseudosavant wrote 14 hours 57 min ago: Not that LLMs are terribly latency sensitive (you wait on a lot of tokens), but what kind of latency impact does this have on requests that go through the proxy? adilhafeez wrote 14 hours 34 min ago: Short answer is latency impact is very minimal. We use envoy as request handler which forwards request to local service written in rust. Envoy is proven to be high performance, low latency and highly efficient on request handling. If I have to put a number it would be in single digit ms per request. I will have more detailed benchmark in the coming days. cotran2 wrote 14 hours 40 min ago: The model is compact 1.5B, most GPUs can serve it locally and has <100ms e2e latency. For L40s, its 50ms. _nh_ wrote 15 hours 51 min ago: How do you compare with RouteLLM? cotran2 wrote 14 hours 41 min ago: There is a case study comparing with RouteLLM in the appendix. sparacha wrote 15 hours 8 min ago: RouteLLM is essentially a benchmark-driven approach. Their framework chooses between a weak and a strong model and helps developers optimize for a metric called APGR (Average Performance Gap Recovered) â a measure of how much of the stronger modelâs performance can be recovered when routing some queries to the weaker, cheaper model. However, their routing models are trained to maximize performance on public benchmarks like MMLU, BBH, or MT-Bench. These benchmarks may not capture subjective, domain-specific quality signals that surface in practice. Arch-Router takes a different approach. Instead of focusing benchmark scores, we lets developers define routing policies in plain language based on their preferences â like âcontract analysis â GPT-4oâ or âlightweight brainstorming â Gemini Flash.â Our 1.5B model learns to map prompts (along with conversational context) to these policies, enabling routing decisions that align with real-world expectations, not abstract leaderboards. Also our approach doesn't require router model retraining when new LLMs are swapped in or when preferences change. Hope this helps. jgant13 wrote 17 hours 28 min ago: Solid. Can you show us when to use this vs. say OpenRouter? The performance seems strong for sure. TIA. sparacha wrote 16 hours 28 min ago: Arch is developer friendly, but designed for enterprise-grade customers in mind. The core contributors of Envoy redesigned the proxy substrate to handle prompts - offering something that is battle tested in terms of resiliency, speed, and deployments. Second, OpenRouter offers choice of models, but dynamically routing to LLMs based on user-defined usage policies is uniquely available in Arch. Hope that helps jedisct1 wrote 17 hours 59 min ago: I tried to use it to rate the difficulty level of coding tasks (for InferSwitch, an LLM router), but it performed far worse than Qwen2.5-Coder-7B (but sure, 1.5B vs 7B) cotran2 wrote 17 hours 33 min ago: According to the post, the model is fine-tuned for routing to different tasks/domains. Classifying difficulty level is probably not the intended use case. sparacha wrote 17 hours 44 min ago: Can you share more about your evaluation setup? I would love to see the specific usage pattern as we have tested our model against smaller LLMs and foundational models and our results show things differently. Of course, routing policies should follow best practices here: [1] Nonetheless, super curious to learn more and see what we may be able to improve. This is technically not a classifier model - its a usage prediction model (feels like a classifier, but not quite in terms of intended usage) URI [1]: https://docs.archgw.com/guides/llm_router.html tmaly wrote 19 hours 17 min ago: do you think it would be possible to quantize this model and still get good results? sparacha wrote 19 hours 15 min ago: yes - we have already published a quantized version here: [1] . The performance difference with a quant version is negligible. I'll run another analysis and update the thread shortly URI [1]: https://huggingface.co/katanemo/Arch-Router-1.5B.gguf sparacha wrote 17 hours 33 min ago: Overall performance degrades from 93.17 -> 92.99 with a quantized version sparacha wrote 21 hours 42 min ago: Hi HN! I am one of the co-authors of the paper. If there are any questions about our approach, I would love to answer them. DIR <- back to front page