DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-14 12:56:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: allow the P-H3-Image text encoder for MiniMax-H3-Image (#2337)

Browse Source
This commit is contained in:
Artem authored and GitHub committed 2026-11-12 23:20:18 +08:00
1 parent 766ce0ec4e
commit 8b1e9df27d
4 files changed
+38 -7
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/conditioning/conditioner.hpp
+12 -3
View File
@@ -2030,11 +2030,20 @@ struct LLMEmbedder : public Conditioner {
"text_encoders.llm",
enable_vision,
weight_manager);
if (version == VERSION_P_H3_IMAGE) {
// MiniMax-H3-Image also accepts the P-H3-Image text encoder in place of the MiniMax-H3 Qwen3-VL-32B encoder.
const bool has_ph3_connector = tensor_storage_map.find(std::string(PH3ImageTextConnectorRunner::prefix) + ".proj.weight") !=
tensor_storage_map.end();
if (version == VERSION_P_H3_IMAGE || (version == VERSION_MINIMAX_H3_IMAGE && has_ph3_connector)) {
if (llm->config.num_layers != 36 || llm->config.hidden_size != 2560) {
throw std::runtime_error("P-H3-Image requires the Qwen3-VL-4B text encoder");
throw std::runtime_error(version == VERSION_P_H3_IMAGE
? "P-H3-Image requires the Qwen3-VL-4B text encoder"
: "The P-H3-Image text connector requires the Qwen3-VL-4B text encoder");
}
ph3_connector = std::make_shared<PH3ImageTextConnectorRunner>(backend, tensor_storage_map, weight_manager);
} else if (version == VERSION_MINIMAX_H3_IMAGE && llm->config.hidden_size == 2560) {
throw std::runtime_error(
"MiniMax-H3-Image with Qwen3-VL-4B requires the P-H3-Image text connector; "
"pass it with --embeddings-connectors or use the MiniMax-H3 Qwen3-VL-32B text encoder");
}
int pad_id = 151643;
if (arch == LLM::LLMArch::MISTRAL_SMALL_3_2 || arch == LLM::LLMArch::MINISTRAL_3_3B) {
@@ -2493,7 +2502,7 @@ struct LLMEmbedder : public Conditioner {
if (sd_version_is_minimax_h3(version)) {
prompt_template_encode_start_idx = 0;
// The final tap is normalized; LLMRunner exposes it at num_layers + 1.
out_layers = version == VERSION_P_H3_IMAGE ? std::set<int>{13, 19, 25, 31, 37} : std::set<int>{50};
out_layers = ph3_connector ? std::set<int>{13, 19, 25, 31, 37} : std::set<int>{50};
prompt_attn_range = {0, 0};
if (llm->enable_vision) {
src/name_conversion.cpp
+1 -1
View File
@@ -1741,7 +1741,7 @@ std::string convert_tensor_name(std::string name, SDVersion version) {
const std::vector<std::pair<std::string, std::string>> connector_prefix_map = {
{"embeddings_connectors.text_encoders.", "text_encoders."},
{"embeddings_connectors.", version == VERSION_P_H3_IMAGE ? "text_encoders.ph3_image_connector." : ""},
{"embeddings_connectors.", sd_version_is_minimax_h3_image(version) ? "text_encoders.ph3_image_connector." : ""},
};
replace_with_prefix_map(name, connector_prefix_map);