deploy: c2d2de4

mit-han-lab · Feb 23, 2024 · 8b683ee · 8b683ee
1 parent 2c95bc3
commit 8b683ee
Show file tree

Hide file tree

Showing 43 changed files with 850 additions and 530 deletions.
diff --git a/Fp32CLIPAttention_8h_source.html b/Fp32CLIPAttention_8h_source.html
@@ -128,11 +128,13 @@
 <div class="line"><a id="l00038" name="l00038"></a><span class="lineno">   38</span>   <span class="keyword">private</span>:</div>
 <div class="line"><a id="l00039" name="l00039"></a><span class="lineno">   39</span>    <span class="keywordtype">void</span> unshape(<a class="code hl_class" href="classMatrix3D.html">Matrix3D&lt;float&gt;</a> shaped, <a class="code hl_class" href="classMatrix3D.html">Matrix3D&lt;float&gt;</a> unshape, <span class="keywordtype">int</span> sqlen);</div>
 <div class="line"><a id="l00040" name="l00040"></a><span class="lineno">   40</span>    <span class="keywordtype">void</span> shape(<a class="code hl_class" href="classMatrix3D.html">Matrix3D&lt;float&gt;</a> unshape, <a class="code hl_class" href="classMatrix3D.html">Matrix3D&lt;float&gt;</a> shaped, <span class="keywordtype">int</span> sqlen);</div>
-<div class="line"><a id="l00041" name="l00041"></a><span class="lineno">   41</span>    <span class="keywordtype">int</span> embed_dim, num_heads, head_dim;</div>
-<div class="line"><a id="l00042" name="l00042"></a><span class="lineno">   42</span>    <a class="code hl_class" href="classLinear__FP.html">Linear_FP</a> k_proj, v_proj, q_proj, out_proj;</div>
-<div class="line"><a id="l00043" name="l00043"></a><span class="lineno">   43</span>    <a class="code hl_class" href="classBMM__F32T.html">BMM_F32T</a> qk_bmm, pv_bmm;</div>
-<div class="line"><a id="l00044" name="l00044"></a><span class="lineno">   44</span>    std::string profile_name = <span class="stringliteral">&quot;Fp32CLIPAttention&quot;</span>;</div>
-<div class="line"><a id="l00045" name="l00045"></a><span class="lineno">   45</span>};</div>
+<div class="line"><a id="l00041" name="l00041"></a><span class="lineno">   41</span>    <span class="comment">// void shape_qkv(Matrix3D&lt;float&gt; unshape, Matrix3D&lt;float&gt; shaped_q, Matrix3D&lt;float&gt; shaped_k,</span></div>
+<div class="line"><a id="l00042" name="l00042"></a><span class="lineno">   42</span>    <span class="comment">//                                       Matrix3D&lt;float&gt; shaped_v, int sqlen);</span></div>
+<div class="line"><a id="l00043" name="l00043"></a><span class="lineno">   43</span>    <span class="keywordtype">int</span> embed_dim, num_heads, head_dim;</div>
+<div class="line"><a id="l00044" name="l00044"></a><span class="lineno">   44</span>    <a class="code hl_class" href="classLinear__FP.html">Linear_FP</a> k_proj, v_proj, q_proj, out_proj, qkv_proj;</div>
+<div class="line"><a id="l00045" name="l00045"></a><span class="lineno">   45</span>    <a class="code hl_class" href="classBMM__F32T.html">BMM_F32T</a> qk_bmm, pv_bmm;</div>
+<div class="line"><a id="l00046" name="l00046"></a><span class="lineno">   46</span>    std::string profile_name = <span class="stringliteral">&quot;Fp32CLIPAttention&quot;</span>;</div>
+<div class="line"><a id="l00047" name="l00047"></a><span class="lineno">   47</span>};</div>
 </div>
 <div class="ttc" id="aclassBMM__F32T_html"><div class="ttname"><a href="classBMM__F32T.html">BMM_F32T</a></div><div class="ttdef"><b>Definition</b> BMM_F32T.h:3</div></div>
 <div class="ttc" id="aclassFp32CLIPAttention_html"><div class="ttname"><a href="classFp32CLIPAttention.html">Fp32CLIPAttention</a></div><div class="ttdef"><b>Definition</b> Fp32CLIPAttention.h:31</div></div>

diff --git a/Fp32CLIPVisionTransformer_8h_source.html b/Fp32CLIPVisionTransformer_8h_source.html
@@ -116,9 +116,9 @@
 <div class="foldopen" id="foldopen00027" data-start="{" data-end="};">
 <div class="line"><a id="l00027" name="l00027"></a><span class="lineno"><a class="line" href="classFp32CLIPVisionTransformer.html">   27</a></span><span class="keyword">class </span><a class="code hl_class" href="classFp32CLIPVisionTransformer.html">Fp32CLIPVisionTransformer</a> {</div>
 <div class="line"><a id="l00028" name="l00028"></a><span class="lineno">   28</span>   <span class="keyword">public</span>:</div>
-<div class="line"><a id="l00029" name="l00029"></a><span class="lineno">   29</span>    <a class="code hl_class" href="classFp32CLIPVisionTransformer.html">Fp32CLIPVisionTransformer</a>(std::string param_path, <span class="keyword">const</span> <span class="keyword">struct</span> <a class="code hl_struct" href="structmodel__config.html">model_config</a> config);</div>
+<div class="line"><a id="l00029" name="l00029"></a><span class="lineno">   29</span>    <a class="code hl_class" href="classFp32CLIPVisionTransformer.html">Fp32CLIPVisionTransformer</a>(std::string param_path, <span class="keyword">const</span> <span class="keyword">struct</span> <a class="code hl_struct" href="structmodel__config.html">model_config</a> config, <span class="keywordtype">bool</span> is_vila);</div>
 <div class="line"><a id="l00030" name="l00030"></a><span class="lineno">   30</span>    <a class="code hl_class" href="classFp32CLIPVisionTransformer.html">Fp32CLIPVisionTransformer</a>(){};</div>
-<div class="line"><a id="l00031" name="l00031"></a><span class="lineno">   31</span>    <span class="keyword">struct </span><a class="code hl_struct" href="structFp32CLIPVisionTransformer__output.html">Fp32CLIPVisionTransformer_output</a> forward(const struct <a class="code hl_struct" href="structFp32CLIPVisionTransformer__input.html">Fp32CLIPVisionTransformer_input</a>&amp; input);</div>
+<div class="line"><a id="l00031" name="l00031"></a><span class="lineno">   31</span>    <span class="keyword">struct </span><a class="code hl_struct" href="structFp32CLIPVisionTransformer__output.html">Fp32CLIPVisionTransformer_output</a> forward(const struct <a class="code hl_struct" href="structFp32CLIPVisionTransformer__input.html">Fp32CLIPVisionTransformer_input</a>&amp; input, <span class="keywordtype">bool</span> is_vila);</div>
 <div class="line"><a id="l00032" name="l00032"></a><span class="lineno">   32</span>    <a class="code hl_class" href="classEmbedding.html">Embedding</a> embed_positions;</div>
 <div class="line"><a id="l00033" name="l00033"></a><span class="lineno">   33</span>    <a class="code hl_class" href="classConv2D.html">Conv2D</a> embed_patch;</div>
 <div class="line"><a id="l00034" name="l00034"></a><span class="lineno">   34</span>    <a class="code hl_class" href="classLayerNorm.html">LayerNorm</a> pre_layernorm;</div>

diff --git a/Generate_8h_source.html b/Generate_8h_source.html
@@ -192,7 +192,7 @@
 <div class="line"><a id="l00101" name="l00101"></a><span class="lineno">  101</span>                             <span class="keyword">const</span> <span class="keyword">struct</span> <a class="code hl_struct" href="structopt__params.html">opt_params</a> generation_config, <a class="code hl_class" href="classEncoder.html">Encoder</a>* encoder = NULL,</div>
 <div class="line"><a id="l00102" name="l00102"></a><span class="lineno">  102</span>                             <span class="keywordtype">bool</span> interactive = <span class="keyword">false</span>, <span class="keywordtype">bool</span> voicechat = <span class="keyword">false</span>);</div>
 <div class="line"><a id="l00103" name="l00103"></a><span class="lineno">  103</span> </div>
-<div class="line"><a id="l00104" name="l00104"></a><span class="lineno">  104</span><span class="keyword">enum</span> { OPT_INT8, LLaMA_FP32, LLaMA_INT4, OPT_FP32, OPT_INT4, StarCoder_FP32, StarCoder_INT4, LLaVA_FP32, LLaVA_INT4 };</div>
+<div class="line"><a id="l00104" name="l00104"></a><span class="lineno">  104</span><span class="keyword">enum</span> { OPT_INT8, LLaMA_FP32, LLaMA_INT4, OPT_FP32, OPT_INT4, StarCoder_FP32, StarCoder_INT4, LLaVA_FP32, LLaVA_INT4, VILA_FP32, VILA_INT4};</div>
 <div class="line"><a id="l00105" name="l00105"></a><span class="lineno">  105</span>std::string LLaMAGenerate(std::string param_path, <span class="keywordtype">void</span>* model, <span class="keywordtype">int</span> model_type, std::string text, <span class="keyword">const</span> <span class="keyword">struct</span> <a class="code hl_struct" href="structopt__params.html">opt_params</a> generation_config,</div>
 <div class="line"><a id="l00106" name="l00106"></a><span class="lineno">  106</span>                          std::string voc_path, <span class="keywordtype">bool</span> interactive, <span class="keywordtype">bool</span> voicechat);</div>
 <div class="line"><a id="l00107" name="l00107"></a><span class="lineno">  107</span> </div>
@@ -201,7 +201,7 @@
 <div class="line"><a id="l00110" name="l00110"></a><span class="lineno">  110</span> </div>
 <div class="line"><a id="l00111" name="l00111"></a><span class="lineno">  111</span>std::string LLaVAGenerate(std::string llama_param_path, <span class="keywordtype">void</span>* llama_model_ptr, std::string clip_param_path, <span class="keywordtype">void</span>* clip_model_ptr, <span class="keywordtype">int</span> model_type, </div>
 <div class="line"><a id="l00112" name="l00112"></a><span class="lineno">  112</span>                          std::string text, std::string img_path, <span class="keyword">const</span> <span class="keyword">struct</span> <a class="code hl_struct" href="structopt__params.html">opt_params</a> generation_config, std::string voc_path, <span class="keywordtype">bool</span> interactive, </div>
-<div class="line"><a id="l00113" name="l00113"></a><span class="lineno">  113</span>                          <span class="keywordtype">bool</span> voicechat);</div>
+<div class="line"><a id="l00113" name="l00113"></a><span class="lineno">  113</span>                          <span class="keywordtype">bool</span> voicechat, <span class="keywordtype">bool</span> is_vila);</div>
 <div class="line"><a id="l00114" name="l00114"></a><span class="lineno">  114</span> </div>
 <div class="line"><a id="l00115" name="l00115"></a><span class="lineno">  115</span><span class="preprocessor">#endif  </span><span class="comment">// GENERATE_H</span></div>
 <div class="ttc" id="aclassEncoder_html"><div class="ttname"><a href="classEncoder.html">Encoder</a></div><div class="ttdef"><b>Definition</b> OPTTokenizer.h:35</div></div>