refactor(tx): fix failed to offload t5

thxCode · thxCode · commit 1e690e1aedde · 2024-12-01T21:00:34.000+08:00
Signed-off-by: thxCode &lt;thxcode0824@gmail.com&gt;
diff --git a/conditioner.hpp b/conditioner.hpp
@@ -1144,7 +1144,7 @@ struct FluxCLIPEmbedder : public Conditioner {
         struct ggml_tensor* pooled              = NULL;  // [768,]
         std::vector<float> hidden_states_vec;
 
-        size_t chunk_len   = 256;
+        size_t chunk_len   = 255;
         size_t chunk_count = t5_tokens.size() / chunk_len;
         for (int chunk_idx = 0; chunk_idx < chunk_count; chunk_idx++) {
             // clip_l
@@ -1231,7 +1231,7 @@ struct FluxCLIPEmbedder : public Conditioner {
                                       int height,
                                       int adm_in_channels        = -1,
                                       bool force_zero_embeddings = false) {
-        auto tokens_and_weights = tokenize(text, 256, true);
+        auto tokens_and_weights = tokenize(text, 255, true);
         return get_learned_condition_common(work_ctx, n_threads, tokens_and_weights, clip_skip, force_zero_embeddings);
     }
 
diff --git a/stable-diffusion.cpp b/stable-diffusion.cpp
@@ -393,15 +393,7 @@ class StableDiffusionGGML {
             first_stage_model->alloc_params_buffer();
             first_stage_model->get_param_tensors(tensors);
         } else {
-            clip_backend   = backend;
-            bool use_t5xxl = false;
-            if (sd_version_is_dit(version)) {
-                use_t5xxl = true;
-            }
-            if (!ggml_backend_is_cpu(backend) && use_t5xxl && conditioner_wtype != GGML_TYPE_F32) {
-                clip_on_cpu = true;
-                LOG_INFO("set clip_on_cpu to true");
-            }
+            clip_backend = backend;
             if (clip_on_cpu && !ggml_backend_is_cpu(backend)) {
                 LOG_INFO("CLIP: Using CPU backend");
                 clip_backend = ggml_backend_cpu_init();
