I switched my local AI setup to AMD's Lemonade after Nvidia support landed, and solved my local AI portability problem
…Model Backend TTFT (ms) TPS VRAM peak (GB) Llama 3.2 3B Vulkan 188 78.3 4.0 Qwen3 8B (Q4_1) ROCm 95 41.2 6.6 Qwen3 8B (Q4_1…
Search
All sources
xda-developers.com
46
huggingface.co
26
blogs.nvidia.com
13
amd.com
13
developer.nvidia.com
5
wccftech.com
2
techpowerup.com
2
aws.amazon.com
1
cnx-software.com
1
hothardware.com
1
androidpolice.com
1
intel.com
1
Tracked topic
Qwen3
Qwen3 is an AI model family developed by Alibaba, released as a set of large language models for natural-language tasks.
Live now
Qwen 3.6 is now much easier to run locally on your Mac, thanks to JetBrains
Deploying Hermes Agent for Free on AMD Developer Cloud with open models and vLLM
…Hermes TUI With the setup complete, your Hermes Agent is now active and powered by the Qwen3.5 model running on your local MI300X. You can now begin interacting with your agent…
Paper page - TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning
…Empirically, TRACE achieves competitive performance and efficiency gains on typical agentic benchmarks, e.g., improving Qwen3-14B Multi-Hop QA average accuracy by 2.8 points over competitive baselines at equal sampling…
I built a local LLM server I can access from anywhere, and it uses a Raspberry Pi
…Just to make things easier for my Raspberry Pi, I swapped back to Qwen 3.5 (0.8B) for the LLM server by running llama-server -hf bartowski/Qwen_Qwen3.5-0…
Paper page - Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVR
…Generated by Qwen/Qwen2.5-Coder-32B-Instruct Reinforcement learning with verifiable rewards (RLVR) has been extended from single-domain training to multi-domain reasoning suites spanning mathematics, programming, and science. However…
Paper page - Tiny but Trusted: Efficient Vision-Language Reasoning for Time-Series Anomaly Detection
…Generated by Qwen/Qwen2.5-Coder-32B-Instruct Recent advances in Vision-Language Models (VLMs) have achieved impressive performance across many tasks, yet prior studies report unsatisfactory performance when applying large language…
Paper page - Wan-Streamer v0.2: Higher Resolution, Same Latency
…The performer becomes a multi-GPU Ulysses-style context-parallel group for the expensive next-unit latent generation. Each performer rank writes incoming K/V into a pre-sharded local cache . The…
Discussions and forums
Ultrafast Qwen3-TTS at 34 ms Time-to-First-Audio, Handling 10 Requests Per Second [OSS]
Hey locallama! We recently open sourced a Qwen3-TTS 1.7B implementation that achieves 10 requests per second (RPS) and sub-50 ms p95 time-to-first-audio (TTFA) while maintaining real-time playback on 1 x H100. This exten…
I'm really hoping we're in 2026's 2-month-gap between QwQ and Qwen3 right now
QwQ was genuine next-gen performance usable on local hardware, but the massive required context (it's reasoning style was akin to "if I say every possible word, I'll notice the right one!") kinda made it unusable for age…
Tesla V100 Qwen3.6 27B Performance
Looking for V100 users to share your config and it's performance. GPU: Tesla V100 PCIE 32Gb Qwen3.6 27B Q4_K_M + Q8_0 MTP 128K context length Pi coding agent llama.cpp model preset: [*] spec-default = 1 ctx-size = 131072…
Qwen3.8-Max matches Kimi K3 and DeepSeek V4 Flash
Qwen3.8-Max (2.4T) is another massive contribution to the open weight community. On benchmarks, it performs closely to Kimi K3 and DeepSeek V4 flash across all categories and is better at coding and software tasks. Qwen3…
I forked Ninfer 3090 and converted it to run on the CMP170HX - doubled my Qwen3.6-35B from llama.cpp
Good afternoon, everyone! I wanted to show the work I've been doing around porting Ninfer over to the CMP170HX (Github) So, first, I do want to call out the amazing work that Neroued, Sergiuszm and specifically Don-Chad …
Paper page - BlockPilot: Instance-Adaptive Policy Learning for Diffusion-based Speculative Decoding
…size varies across samples and plays a critical role in speculative decoding performance. Moreover, these values exhibit a clear local structure, concentrating around the training block size, which reduces the problem to…
Paper page - Dr-DCI: Scaling Direct Corpus Interaction via Dynamic Workspace Expansion
…The agent first pulls potentially relevant documents into its workspace, then performs local DCI operations — search, filter, compare, inspect, and verify — within that evolving workspace. 💡 Why it works: DR-DCI combines the…
Paper page - ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
…an expert-locality-aware decode router for prefill-decode disaggregated Mixture-of-Experts serving that improves performance by predicting expert activations and routing requests accordingly. Generated by Qwen/Qwen2.5-Coder-32B…