News:

Willkommen im Notebookcheck.com Forum! Hier können Sie über alle unsere Artikel und allgemein über notebookrelevante Dinge diskutieren. Viel Spass!

Main Menu

Post reply

Other options
Verification:
Please leave this box empty:
Shortcuts: ALT+S post or ALT+P preview

Topic summary

Posted by DeepSeek-V4-Flash-0731
 - Today at 12:57:42
Open-weight models, that is.
Posted by DeepSeek-V4-Flash-0731
 - Today at 12:56:35
See various up to around 300B parameter AI LLM models compared:
artificialanalysis.ai/?models=qwen3-5-122b-a10b-non-reasoning%2Cqwen3-5-122b-a10b%2Cmimo-v2-5-0424%2Cqwen3-6-27b-non-reasoning%2Cminimax-m2-7%2Cnvidia-nemotron-3-super-120b-a12b%2Cqwen3-6-35b-a3b-non-reasoning%2Cstep-3-7-flash%2Cqwen3-6-35b-a3b%2Cqwen3-6-27b%2Cmotif-0714%2Chy3%2Cgemma-4-31b%2Cgemma-4-31b-non-reasoning%2Cmistral-medium-3-5%2Cgpt-oss-120b%2Cdeepseek-v4-flash%2Cinkling-small&intelligence=artificial-analysis-intelligence-index
Posted by DeepSeek-V4-Flash-0731
 - Today at 11:37:51
Quote from: 192 GB+24 GB VRAM is nice on July 21, 2026, 11:06:14(the current DeekSeek-V4-Flash is a preview version and the final version is supposed so be released this month)
And, as said, the full version has been released this very month:
Quote from: reddit.com/r/LocalLLaMA/comments/1vbidkp/deepseekv4flash_has_been_updated_the_official/DeepSeek-V4-Flash has been updated, "The official release of DeepSeek-V4-Pro will follow soon"
Quote from: reddit.com/r/LocalLLaMA/comments/1vbk5ob/new_deepseek_v4flash_achieves_50_on/New DeepSeek V4-Flash achieves 50 on ArtificalAnalysis Index, 1 point below GLM-5.2 and GPT-5.6 Luna
Ok, I don't think the weights have been released yet.

On 128 GB RAM systems, like Strix Halo (note: AFAIK, on Linux more RAM can be allocated), it may be worth trying to run a 3-bit quant (smallest is 103 GB, huggingface.co/unsloth/DeepSeek-V4-Flash-GGUF).

And the upcoming Gorgon Halo 192 GB RAM configs will fit the full DeepSeek V4 Flash weights (162 GB) + have plenty of memory left for the context on Linux, where more RAM can be allocated than on Windows (if still so), just like with Strix Halo.

Meanwhile llama.cpp DeepSeek-V4 support is coming along nicely, too:
llama.cpp] git log --pretty=format:"%h %ad %s" --date=short -P --grep "deepseek(?:.*?[^[:alnum:]])?v?4\b|ds-?v? *?[-+_]?v?4" -i
9d9a6d29f 2026-07-31 SYCL: add oneMKL GEMM flash attention for XMX-accelerated prompt proc... (#25025)
69e62fc77 2026-07-31 llama : enforce the same K and V cache types for DeepSeek V4; enable FA if V cache is quantized (#25871)
dee2a846b 2026-07-27 ggml : adjust logic for offloading ops to weight's backend (#25832)
e8e6c7af2 2026-07-22 minor: fix reasoning preserve var for DS4 [no ci] (#25999)
f534da26e 2026-07-22 Fix DeepSeek4 crafted template (#25414)
67b9b0e7f 2026-07-22 llama-arch: fix DeepSeek4 APE tensor op (#25945)
846e991ec 2026-07-22 cuda: add sqrt_softplus in topk-moe for dsv4 (#25896)
91d2fc387 2026-07-20 llama_dsv4: write only used rows in state (#25325)
4937ca83f 2026-07-18 llama-quant : exclude i32 ffn_gate_tid2eid routing table from quantization (#25787)
5d5306bf3 2026-07-17 tests : initialize all tensors in test_dsv4_hc to avoid NaNs in sentinel tensors (#25822)
0dc74e332 2026-07-17 DeepseekV4: Add fused hyper-connection ops (#25585)
33a75f41c 2026-07-15 DeepseekV4: reduce graph splits (#25702)
7f575c39d 2026-07-14 DeepseekV4: fix seq_rm (#25588)
13f2b28b0 2026-07-11 DeepseekV4: clear cache only for seq rather than full (#25521)
00f5442cc 2026-07-11 ggml : add GGML_OP_LIGHTNING_INDEXER that implements DeepSeek V3.2/V4 lightning indexer (#24231)
2ed3c1abb 2026-07-10 llama : make all KQ masks f16 if FA is used, remove zero attention bias, remove raw_k repeats in DeepSeek V4 (#25370)
024c46ae4 2026-07-07 llama: fix quantized kv-cache for dsv4 (#25202)
8c146a836 2026-06-29 DeepSeek V4  (#24162)