GithubStar/ktransformers
1
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2026-10-17 05:11:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

deploy: 4882505c9a

Browse Source
This commit is contained in:
yyj6666667
2026-09-13 13:52:17 +00:00
parent 667db91084
commit 8820558b4f
24 changed files with 31 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
404.html
+1 -1
View File
@@ -36,7 +36,7 @@
const path_to_root = "";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "searchindex-8e125c1e.js";
window.path_to_searchindex_js = "searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="toc-78d73750.js"></script>
en/FAQ.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../toc-78d73750.js"></script>
en/SFT/DPO_tutorial.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/SFT/KTransformers-Fine-Tuning_Cookbook.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/SFT/KTransformers-Fine-Tuning_Developer-Technical-Notes.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/SFT/KTransformers-Fine-Tuning_User-Guide.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/SFT/Qwen3.5-SGLang-LoRA-Serving.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/SFT/index.html
+2 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
@@ -182,6 +182,7 @@
<main>
<h1 id="ktransformers-sft-docs"><a class="header" href="#ktransformers-sft-docs">KTransformers SFT Docs</a></h1>
<ul>
<li><a href="../../../.github/release/examples/kimi-k25/README.html">Kimi K2.5: PyPI Install, LoRA Training, Resume and Serving (中文)</a></li>
<li><a href="./KTransformers-Fine-Tuning_Cookbook.html">KTransformers × LlamaFactory MoE Fine-Tuning Cookbook</a></li>
<li><a href="../../zh/KTransformers-Fine-Tuning_Cookbook_zh.html">KTransformers × LlamaFactory MoE Fine-Tuning Cookbook (中文)</a></li>
<li><a href="./KTransformers-Fine-Tuning_Quick-Start.html">v0.6.1 Quick Start</a></li>
en/SFT/injection_tutorial.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/kt-kernel/AVX2-Tutorial.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/kt-kernel/GLM-5.3-Flash-Tutorial.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/kt-kernel/index.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/kt-kernel/kt-cli.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
en/kt-kernel/kt-kernel_intro.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../../toc-78d73750.js"></script>
index.html
+4 -2
View File
@@ -35,7 +35,7 @@
const path_to_root = "";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "searchindex-8e125c1e.js";
window.path_to_searchindex_js = "searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="toc-78d73750.js"></script>
@@ -198,11 +198,12 @@
<p>KTransformers is a research project focused on efficient inference and fine-tuning of large language models through CPU-GPU heterogeneous computing. The project now exposes two user-facing capabilities from the kt-kernel source tree: <a href="./kt-kernel/README.html">Inference</a> and <a href="./doc/en/SFT/KTransformers-Fine-Tuning_Cookbook.html">SFT</a>.</p>
<h2 id="-updates"><a class="header" href="#-updates">🔥 Updates</a></h2>
<ul>
<li><strong>Sep 13, 2026</strong>: <strong>Kimi K2.5 / K2.6 LoRA fine-tuning</strong> is supported with native RAWINT4 routed experts.</li>
<li><strong>Aug 26, 2026</strong>: Added native support for <strong>GLM-5.3-flash</strong>, bringing 1M-token context and multimodal input to consumer GPUs. (<a href="./doc/en/kt-kernel/GLM-5.3-Flash-Tutorial.html">Tutorial</a>)</li>
<li><strong>Aug 25, 2026</strong>: Uploaded a new easy-to-use <a href="./doc/en/SFT/KTransformers-Fine-Tuning_Cookbook.html">KTransformers × LlamaFactory MoE Fine-Tuning Cookbook</a>, covering hardware checks, installation, BF16/FP8/INT8 recipes, LoRA and full fine-tuning, resource planning, and troubleshooting.</li>
<li><strong>Aug 17, 2026</strong>: LoRA fine-tuning now supports compatible AVX512 x86 CPUs, including AMD servers, without requiring AMX. (<a href="https://github.com/kvcache-ai/ktransformers/releases/tag/v0.7.0">v0.7.0 Release Notes</a>)</li>
<li><strong>Aug 16, 2026</strong>: DeepSeek-V4-Flash on a single Ascend NPU with CPU expert offload. (<a href="./doc/en/DeepSeek-V4-Flash_tutorial_for_Ascend_NPU.html">Tutorial</a>)</li>
<li><strong>Aug 5, 2026</strong>: Introduced native block-FP8 LoRA fine-tuning, loading FP8 routed-expert weights directly from the checkpoint without materializing a complete BF16 copy. (<a href="https://github.com/kvcache-ai/ktransformers/pull/2141">PR #2141</a>)</li>
<li><strong>Aug 5, 2026</strong>: Introduced native block-FP8 LoRA fine-tuning (<a href="https://github.com/kvcache-ai/ktransformers/pull/2141">PR #2141</a>)</li>
<li><strong>July 23, 2026</strong>: Added end-to-end BF16 full-parameter fine-tuning for MoE models, including complete checkpoint saving. (<a href="https://github.com/kvcache-ai/ktransformers/pull/2094">PR #2094</a>)</li>
<li><strong>June 21, 2026</strong>: MiniMax-M3 Day0 Support! (<a href="./doc/en/kt-kernel/MiniMax-M3-Tutorial.html">Tutorial</a>)</li>
<li><strong>June 17, 2026</strong>: GLM-5.2 Day0 Support! (<a href="./doc/en/kt-kernel/GLM-5.2-Tutorial.html">Tutorial</a>)</li>
@@ -278,6 +279,7 @@ pip install .
<hr>
<h3 id="-sft---fine-tuning-with-llamafactory"><a class="header" href="#-sft---fine-tuning-with-llamafactory">🎓 <a href="./doc/en/SFT/KTransformers-Fine-Tuning_Cookbook.html">SFT</a> - Fine-Tuning with LlamaFactory</a></h3>
<p>KTransformers × LlamaFactory integration for ultra-large MoE model fine-tuning. The new Cookbook provides an easy-to-use path from hardware checks and installation to BF16/FP8/INT8 configuration, LoRA/full training, resource planning, and troubleshooting.</p>
<p><strong>Kimi K2.5 / K2.6 LoRA fine-tuning</strong> is supported with native RAWINT4 routed experts. Follow the <a href="./.github/release/examples/kimi-k25/README.html">PyPI installation, training, resume and SGLang serving guide (中文)</a>. End-to-end validation uses Kimi K2.5; K2.6 follows the same model-architecture path.</p>
<p><label class="checkbox-label"><input class="checkbox-img" type="checkbox"><img src="./doc/assets/llamafactory-kt-integration-overview.png" alt="LlamaFactory and KTransformers MoE fine-tuning architecture"><span class="img-wrapper"><img src="./doc/assets/llamafactory-kt-integration-overview.png" alt="LlamaFactory and KTransformers MoE fine-tuning architecture"></span></label></p>
<p><strong>Key Features:</strong></p>
<ul>
print.html
+5 -2
View File
@@ -36,7 +36,7 @@
const path_to_root = "";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "searchindex-8e125c1e.js";
window.path_to_searchindex_js = "searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="toc-78d73750.js"></script>
@@ -197,11 +197,12 @@
<p>KTransformers is a research project focused on efficient inference and fine-tuning of large language models through CPU-GPU heterogeneous computing. The project now exposes two user-facing capabilities from the kt-kernel source tree: <a href="kt-kernel/README.html">Inference</a> and <a href="doc/en/SFT/KTransformers-Fine-Tuning_Cookbook.html">SFT</a>.</p>
<h2 id="-updates"><a class="header" href="#-updates">🔥 Updates</a></h2>
<ul>
<li><strong>Sep 13, 2026</strong>: <strong>Kimi K2.5 / K2.6 LoRA fine-tuning</strong> is supported with native RAWINT4 routed experts.</li>
<li><strong>Aug 26, 2026</strong>: Added native support for <strong>GLM-5.3-flash</strong>, bringing 1M-token context and multimodal input to consumer GPUs. (<a href="doc/en/kt-kernel/GLM-5.3-Flash-Tutorial.html">Tutorial</a>)</li>
<li><strong>Aug 25, 2026</strong>: Uploaded a new easy-to-use <a href="doc/en/SFT/KTransformers-Fine-Tuning_Cookbook.html">KTransformers × LlamaFactory MoE Fine-Tuning Cookbook</a>, covering hardware checks, installation, BF16/FP8/INT8 recipes, LoRA and full fine-tuning, resource planning, and troubleshooting.</li>
<li><strong>Aug 17, 2026</strong>: LoRA fine-tuning now supports compatible AVX512 x86 CPUs, including AMD servers, without requiring AMX. (<a href="https://github.com/kvcache-ai/ktransformers/releases/tag/v0.7.0">v0.7.0 Release Notes</a>)</li>
<li><strong>Aug 16, 2026</strong>: DeepSeek-V4-Flash on a single Ascend NPU with CPU expert offload. (<a href="doc/en/DeepSeek-V4-Flash_tutorial_for_Ascend_NPU.html">Tutorial</a>)</li>
<li><strong>Aug 5, 2026</strong>: Introduced native block-FP8 LoRA fine-tuning, loading FP8 routed-expert weights directly from the checkpoint without materializing a complete BF16 copy. (<a href="https://github.com/kvcache-ai/ktransformers/pull/2141">PR #2141</a>)</li>
<li><strong>Aug 5, 2026</strong>: Introduced native block-FP8 LoRA fine-tuning (<a href="https://github.com/kvcache-ai/ktransformers/pull/2141">PR #2141</a>)</li>
<li><strong>July 23, 2026</strong>: Added end-to-end BF16 full-parameter fine-tuning for MoE models, including complete checkpoint saving. (<a href="https://github.com/kvcache-ai/ktransformers/pull/2094">PR #2094</a>)</li>
<li><strong>June 21, 2026</strong>: MiniMax-M3 Day0 Support! (<a href="doc/en/kt-kernel/MiniMax-M3-Tutorial.html">Tutorial</a>)</li>
<li><strong>June 17, 2026</strong>: GLM-5.2 Day0 Support! (<a href="doc/en/kt-kernel/GLM-5.2-Tutorial.html">Tutorial</a>)</li>
@@ -277,6 +278,7 @@ pip install .
<hr>
<h3 id="-sft---fine-tuning-with-llamafactory"><a class="header" href="#-sft---fine-tuning-with-llamafactory">🎓 <a href="doc/en/SFT/KTransformers-Fine-Tuning_Cookbook.html">SFT</a> - Fine-Tuning with LlamaFactory</a></h3>
<p>KTransformers × LlamaFactory integration for ultra-large MoE model fine-tuning. The new Cookbook provides an easy-to-use path from hardware checks and installation to BF16/FP8/INT8 configuration, LoRA/full training, resource planning, and troubleshooting.</p>
<p><strong>Kimi K2.5 / K2.6 LoRA fine-tuning</strong> is supported with native RAWINT4 routed experts. Follow the <a href=".github/release/examples/kimi-k25/README.html">PyPI installation, training, resume and SGLang serving guide (中文)</a>. End-to-end validation uses Kimi K2.5; K2.6 follows the same model-architecture path.</p>
<p><label class="checkbox-label"><input class="checkbox-img" type="checkbox"><img src="doc/assets/llamafactory-kt-integration-overview.png" alt="LlamaFactory and KTransformers MoE fine-tuning architecture"><span class="img-wrapper"><img src="doc/assets/llamafactory-kt-integration-overview.png" alt="LlamaFactory and KTransformers MoE fine-tuning architecture"></span></label></p>
<p><strong>Key Features:</strong></p>
<ul>
@@ -1355,6 +1357,7 @@ chunk_size: 8192
<div style="break-before: page; page-break-before: always;"></div>
<h1 id="ktransformers-sft-docs"><a class="header" href="#ktransformers-sft-docs">KTransformers SFT Docs</a></h1>
<ul>
<li><a href="../.github/release/examples/kimi-k25/README.html">Kimi K2.5: PyPI Install, LoRA Training, Resume and Serving (中文)</a></li>
<li><a href="#ktransformers--llamafactory-moe-fine-tuning-cookbook">KTransformers × LlamaFactory MoE Fine-Tuning Cookbook</a></li>
<li><a href="#ktransformers--llamafactory-moe-微调-cookbook">KTransformers × LlamaFactory MoE Fine-Tuning Cookbook (中文)</a></li>
<li><a href="en/SFT/KTransformers-Fine-Tuning_Quick-Start.html">v0.6.1 Quick Start</a></li>
searcher-09f2665d.js
+1 -1
View File
@@ -440,7 +440,7 @@ window.search = window.search || {};
if (yes) {
loadSearchScript(
window.path_to_searchindex_js ||
path_to_root + 'searchindex-8e125c1e.js',
path_to_root + 'searchindex-a0ca3e17.js',
'mdbook-search-index');
search_wrap.classList.remove('hidden');
searchicon.setAttribute('aria-expanded', 'true');
searchindex-8e125c1e.js
-1
View File
File diff suppressed because one or more lines are too long
searchindex-a0ca3e17.js
+1
View File
File diff suppressed because one or more lines are too long
zh/AVX2-Tutorial_zh.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../toc-78d73750.js"></script>
zh/GLM-5.3-Flash-Tutorial_zh.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../toc-78d73750.js"></script>
zh/KTransformers-Fine-Tuning_Cookbook_zh.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../toc-78d73750.js"></script>
zh/Qwen3.5-SGLang-LoRA-Serving_zh.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../toc-78d73750.js"></script>
zh/SYCL-GPTQ-INT4-Tutorial_zh.html
+1 -1
View File
@@ -35,7 +35,7 @@
const path_to_root = "../";
const default_light_theme = "light";
const default_dark_theme = "navy";
window.path_to_searchindex_js = "../searchindex-8e125c1e.js";
window.path_to_searchindex_js = "../searchindex-a0ca3e17.js";
</script>
<!-- Start loading toc.js asap -->
<script src="../toc-78d73750.js"></script>