vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix embedding layer based on Noeda's example

Browse source
This commit is contained in:
S 2024-04-05 10:35:19 +01:00
parent c354db751e
commit 553b09ba8f
2 changed files with 34 additions and 59 deletions
Download patch file Download diff file Expand all files Collapse all files

48
convert-hf-to-gguf.py
View file

@ -2352,54 +2352,6 @@ class CommandR2Model(Model):
self.gguf_writer.add_logit_scale(self.hparams["logit_scale"])
self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.NONE)
def write_tensors(self):
block_count = self.hparams.get("n_layers", self.hparams.get("num_hidden_layers", self.hparams.get("n_layer")))
tensor_map = gguf.get_tensor_name_map(self.model_arch, block_count)
for name, data_torch in self.get_tensors():
# we don't need these
if name.endswith((".attention.masked_bias", ".attention.bias", ".attention.rotary_emb.inv_freq")):
continue
#Convert Q norm and K norm to 1d so they are exported in float32 and not quantized
if name.endswith((".q_norm.weight")) or name.endswith((".k_norm.weight")):
data_torch = data_torch.flatten()
old_dtype = data_torch.dtype
# convert any unsupported data types to float32
if data_torch.dtype not in (torch.float16, torch.float32):
data_torch = data_torch.to(torch.float32)
data = data_torch.squeeze().numpy()
# map tensor names
new_name = tensor_map.get_name(name, try_suffixes=(".weight", ".bias"))
if new_name is None:
print(f"Can not map tensor {name!r}")
sys.exit()
n_dims = len(data.shape)
data_dtype = data.dtype
# if f32 desired, convert any float16 to float32
if self.ftype == 0 and data_dtype == np.float16:
data = data.astype(np.float32)
# TODO: Why cant we use these float16 as-is? There should be not reason to store float16 as float32
if self.ftype == 1 and data_dtype == np.float16 and n_dims == 1:
data = data.astype(np.float32)
# if f16 desired, convert any float32 2-dim weight tensors to float16
if self.ftype == 1 and data_dtype == np.float32 and name.endswith(".weight") and n_dims == 2:
data = data.astype(np.float16)
print(f"{new_name}, n_dims = {n_dims}, {old_dtype} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data)
###### CONVERSION LOGIC ######

29
llama.cpp
View file

@ -5408,8 +5408,8 @@ static bool llm_load_tensors(
if (n_layer >= 64)
{
layer.attn_q_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {hparams.n_embd_head_k * hparams.n_head});
layer.attn_k_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {hparams.n_embd_head_k * hparams.n_head_kv});
layer.attn_q_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {hparams.n_embd_head_k, hparams.n_head});
layer.attn_k_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {hparams.n_embd_head_k, hparams.n_head_kv});
}
layer.wq = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd});
@ -9462,6 +9462,18 @@ struct llm_build_context {
if (model.layers[il].attn_q_norm)
{
Qcur = ggml_view_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens,
ggml_element_size(Qcur) * n_embd_head,
ggml_element_size(Qcur) * n_embd_head * n_head,
0);
cb(Qcur, "Qcur", il);
Kcur = ggml_view_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens,
ggml_element_size(Kcur) * n_embd_head,
ggml_element_size(Kcur) * n_embd_head * n_head_kv,
0);
cb(Kcur, "Kcur", il);
Qcur = llm_build_norm(ctx0, Qcur, hparams,
model.layers[il].attn_q_norm,
NULL,
@ -13085,9 +13097,15 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
return std::make_pair(i_layer, n_layer);
};
// Command-R+ has such a large embedding weight tensor it overflows
// 32-bit signed integers. This is band-aid until quants can deal with
// that.
if (name == "token_embd.weight" && arch == LLM_ARCH_COMMAND_R && qs.model.hparams.n_layer >= 64) {
new_type = GGML_TYPE_F16;
}
// for arches that share the same tensor between the token embeddings and the output, we quantize the token embeddings
// with the quantization of the output tensor
if (name == tn(LLM_TENSOR_OUTPUT, "weight") || (!qs.has_output && name == tn(LLM_TENSOR_TOKEN_EMBD, "weight"))) {
else if (name == tn(LLM_TENSOR_OUTPUT, "weight") || (!qs.has_output && name == tn(LLM_TENSOR_TOKEN_EMBD, "weight"))) {
if (qs.params->output_tensor_type < GGML_TYPE_COUNT) {
new_type = qs.params->output_tensor_type;
} else {
@ -13119,6 +13137,11 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
new_type = GGML_TYPE_IQ3_S;
}
}
} else if ((arch == LLM_ARCH_COMMAND_R) &&
(name.find("q_norm") != std::string::npos ||
name.find("k_norm") != std::string::npos)) {
new_type = GGML_TYPE_F32;
} else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS || ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ1_S ||
ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M || ftype == LLAMA_FTYPE_MOSTLY_IQ1_M) {
if (name.find("attn_v.weight") != std::string::npos) {