Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 09:15:18 +02:00
Code Issues Packages Projects Releases Wiki Activity

wip qwen image edit. not working yet

Browse Source
This commit is contained in:
Concedo
2025-10-11 11:24:17 +08:00
parent 6d8f8cd65b
commit 80f88eb703
17 changed files with 1482 additions and 432 deletions
Download Patch File Download Diff File Expand all files Collapse all files
otherarch/sdcpp/model.cpp
+39 -8
View File
@@ -129,7 +129,6 @@ const char* unused_tensors[] = {
"text_encoders.t5xxl.transformer.encoder.embed_tokens.weight", // only used during training
"text_encoders.qwen2vl.output.weight",
"text_encoders.qwen2vl.lm_head.",
"text_encoders.qwen2vl.visual.",
};
bool is_unused_tensor(std::string name) {
@@ -228,6 +227,24 @@ std::unordered_map<std::string, std::string> qwenvl_name_map{
{"output_norm.", "model.norm."},
};
std::unordered_map<std::string, std::string> qwenvl_vision_name_map{
{"mm.", "merger.mlp."},
{"v.post_ln.", "merger.ln_q."},
{"v.patch_embd.weight", "patch_embed.proj.0.weight"},
{"patch_embed.proj.0.weight.1", "patch_embed.proj.1.weight"},
{"v.patch_embd.weight.1", "patch_embed.proj.1.weight"},
{"v.blk.", "blocks."},
{"attn_q.", "attn.q_proj."},
{"attn_k.", "attn.k_proj."},
{"attn_v.", "attn.v_proj."},
{"attn_out.", "attn.proj."},
{"ffn_down.", "mlp.down_proj."},
{"ffn_gate.", "mlp.gate_proj."},
{"ffn_up.", "mlp.up_proj."},
{"ln1.", "norm1."},
{"ln2.", "norm2."},
};
std::string convert_cond_model_name(const std::string& name) {
std::string new_name = name;
std::string prefix;
@@ -286,10 +303,19 @@ std::string convert_cond_model_name(const std::string& name) {
new_name.replace(pos, 11, "layer.0.SelfAttention.relative_attention_bias.");
}
} else if (contains(name, "qwen2vl")) {
for (auto kv : qwenvl_name_map) {
size_t pos = new_name.find(kv.first);
if (pos != std::string::npos) {
new_name.replace(pos, kv.first.size(), kv.second);
if (contains(name, "qwen2vl.visual")) {
for (auto kv : qwenvl_vision_name_map) {
size_t pos = new_name.find(kv.first);
if (pos != std::string::npos) {
new_name.replace(pos, kv.first.size(), kv.second);
}
}
} else {
for (auto kv : qwenvl_name_map) {
size_t pos = new_name.find(kv.first);
if (pos != std::string::npos) {
new_name.replace(pos, kv.first.size(), kv.second);
}
}
}
} else if (name == "text_encoders.t5xxl.transformer.token_embd.weight") {
@@ -1879,10 +1905,15 @@ SDVersion ModelLoader::get_sd_version() {
}
if (is_flux) {
is_inpaint = input_block_weight.ne[0] == 384;
if (is_inpaint) {
if (input_block_weight.ne[0] == 384) {
return VERSION_FLUX_FILL;
}
if (input_block_weight.ne[0] == 128) {
return VERSION_FLUX_CONTROLS;
}
if (input_block_weight.ne[0] == 196) {
return VERSION_FLEX_2;
}
return VERSION_FLUX;
}
@@ -2637,4 +2668,4 @@ bool convert(const char* input_path, const char* vae_path, const char* output_pa
}
bool success = model_loader.save_to_gguf_file(output_path, (ggml_type)output_type, tensor_type_rules);
return success;
}
}