mtmd: add MiMo v2.5 vision (#22883)
* mimo-v2.5: vision support * mimo-v2.5: use fused qkv for vision * mimi-v2.5: fix f16 vision overflow * mimo-v2.5: comment cleanups * mimo-v2.5: Flash doesn't have mmproj more cleanup remember to use filter_tensors * mimo-v2.5: fix trailing whitespace
This commit is contained in:
@@ -31,6 +31,7 @@
|
||||
#define KEY_N_BLOCK "clip.%s.block_count"
|
||||
#define KEY_PROJ_DIM "clip.%s.projection_dim"
|
||||
#define KEY_N_HEAD "clip.%s.attention.head_count"
|
||||
#define KEY_N_HEAD_KV "clip.%s.attention.head_count_kv"
|
||||
#define KEY_LAYER_NORM_EPS "clip.%s.attention.layer_norm_epsilon"
|
||||
|
||||
// vision-specific
|
||||
@@ -53,6 +54,7 @@
|
||||
#define KEY_IMAGE_GRID_PINPOINTS "clip.vision.image_grid_pinpoints"
|
||||
#define KEY_WIN_ATTN_PATTERN "clip.vision.n_wa_pattern"
|
||||
#define KEY_WIN_ATTN_LAYER_INDEXES "clip.vision.wa_layer_indexes"
|
||||
#define KEY_WA_PATTERN_MODE "clip.vision.wa_pattern_mode"
|
||||
#define KEY_ATTN_WINDOW_SIZE "clip.vision.window_size"
|
||||
#define KEY_MINICPMV_VERSION "clip.minicpmv_version"
|
||||
#define KEY_MINICPMV_QUERY_NUM "clip.minicpmv_query_num"
|
||||
@@ -86,6 +88,7 @@
|
||||
#define TN_ATTN_Q "%s.blk.%d.attn_q.%s"
|
||||
#define TN_ATTN_V "%s.blk.%d.attn_v.%s"
|
||||
#define TN_ATTN_OUTPUT "%s.blk.%d.attn_out.%s"
|
||||
#define TN_ATTN_SINKS "%s.blk.%d.attn_sinks"
|
||||
#define TN_ATTN_K_NORM "%s.blk.%d.attn_k_norm.%s"
|
||||
#define TN_ATTN_Q_NORM "%s.blk.%d.attn_q_norm.%s"
|
||||
#define TN_FFN_DOWN "%s.blk.%d.ffn_down.%s"
|
||||
@@ -344,6 +347,7 @@ enum projector_type {
|
||||
PROJECTOR_TYPE_HUNYUANVL,
|
||||
PROJECTOR_TYPE_MINICPMV4_6,
|
||||
PROJECTOR_TYPE_GRANITE_SPEECH,
|
||||
PROJECTOR_TYPE_MIMOVL,
|
||||
PROJECTOR_TYPE_UNKNOWN,
|
||||
};
|
||||
|
||||
@@ -393,6 +397,7 @@ static std::map<projector_type, std::string> PROJECTOR_TYPE_NAMES = {
|
||||
{ PROJECTOR_TYPE_HUNYUANVL, "hunyuanvl"},
|
||||
{ PROJECTOR_TYPE_MINICPMV4_6, "minicpmv4_6"},
|
||||
{ PROJECTOR_TYPE_GRANITE_SPEECH, "granite_speech"},
|
||||
{ PROJECTOR_TYPE_MIMOVL, "mimovl"},
|
||||
};
|
||||
|
||||
static projector_type clip_projector_type_from_string(const std::string & str) {
|
||||
|
||||
Reference in New Issue
Block a user