Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat: add changes to handle jina v2 base code #7596

Merged
merged 10 commits into from
Jun 6, 2024
Merged

feat: add changes to handle jina v2 base code #7596

Show file tree
Hide file tree
Changes from 7 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
cc0ac09
feat: add changes to handle jina v2 base code
JoanFM May 28, 2024
21936dd
fix: do not complicate things
JoanFM May 28, 2024
9a65c7a
fix: fix the usage of the code model
JoanFM May 31, 2024
96a6f55
Merge branch 'master' of https://github.com/JoanFM/llama.cpp into fea…
JoanFM May 31, 2024
0fc775e
Merge branch 'master' of https://github.com/JoanFM/llama.cpp into fea…
JoanFM Jun 4, 2024
4bce30c
fix: fix comments
JoanFM Jun 4, 2024
3b44f8f
fix: fix linting issues
JoanFM Jun 5, 2024
05659d3
fix: remove ollama patches
JoanFM Jun 5, 2024
7ab6023
Merge branch 'master' of https://github.com/JoanFM/llama.cpp into fea…
JoanFM Jun 5, 2024
4c4d877
style : minor
ggerganov Jun 6, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions convert-hf-to-gguf-update.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,6 +82,7 @@ class TOKENIZER_TYPE(IntEnum):
{"name": "jina-v2-es", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-es", },
{"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-de", },
{"name": "smaug-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", },
{"name": "jina-v2-code", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", },
]


Expand Down
7 changes: 6 additions & 1 deletion convert-hf-to-gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -474,6 +474,9 @@ def get_vocab_base_pre(self, tokenizer) -> str:
if chkhsh == "c136ed14d01c2745d4f60a9596ae66800e2b61fa45643e72436041855ad4089d":
# ref: https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct
res = "smaug-bpe"
if chkhsh == "7967bfa498ade6b757b064f31e964dddbb80f8f9a4d68d4ba7998fcf281c531a":
# ref: https://huggingface.co/jinaai/jina-embeddings-v2-base-code
res = "jina-v2-code"

if res is None:
logger.warning("\n")
Expand Down Expand Up @@ -2451,11 +2454,13 @@ def __init__(self, *args, **kwargs):

def get_tensors(self):
for name, data in super().get_tensors():
if 'gated_layers' in name:
if 'gated_layer' in name:
d1 = data[:self.intermediate_size, :]
name1 = name.replace('gated_layers', 'gated_layers_w')
name1 = name1.replace('up_gated_layer', 'gated_layers_v')
d2 = data[self.intermediate_size:, :]
name2 = name.replace('gated_layers', 'gated_layers_v')
name2 = name2.replace('up_gated_layer', 'gated_layers_w')
yield name1, d1
yield name2, d2
continue
Expand Down
1 change: 1 addition & 0 deletions gguf-py/gguf/constants.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -426,6 +426,7 @@ class MODEL_TENSOR(IntEnum):
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.LAYER_OUT_NORM,
MODEL_TENSOR.ATTN_NORM_2,
],
MODEL_ARCH.MPT: [
MODEL_TENSOR.TOKEN_EMBD,
Expand Down
3 changes: 3 additions & 0 deletions gguf-py/gguf/tensor_mapping.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,6 +102,7 @@ class TensorNameMap:
# Attention norm 2
MODEL_TENSOR.ATTN_NORM_2: (
"transformer.h.{bid}.ln_attn", # falcon40b
"encoder.layer.{bid}.layer_norm_1", # jina-v2-code
),

# Attention query-key-value
Expand Down Expand Up @@ -311,6 +312,7 @@ class TensorNameMap:
"model.layers.{bid}.mlp.c_proj", # starcoder2
"encoder.layer.{bid}.mlp.wo", # jina-bert-v2
"model.layers.{bid}.residual_mlp.w2", # arctic
"encoder.layer.{bid}.mlp.down_layer", # jina-bert-v2
),

MODEL_TENSOR.FFN_DOWN_EXP: (
Expand Down Expand Up @@ -350,6 +352,7 @@ class TensorNameMap:
"encoder.layers.{bid}.norm2", # nomic-bert
"transformer.decoder_layer.{bid}.rms_norm_3", # Grok
"encoder.layer.{bid}.mlp.layernorm", # jina-bert-v2
"encoder.layer.{bid}.layer_norm_2" # jina-v2-code
),

MODEL_TENSOR.SSM_IN: (
Expand Down
54 changes: 32 additions & 22 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -717,6 +717,7 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
{ LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" },
{ LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" },
{ LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" },
{ LLM_TENSOR_ATTN_NORM_2, "blk.%d.attn_norm_2" },
},
},
{
Expand Down Expand Up @@ -4652,16 +4653,7 @@ static void llm_load_vocab(

// for now, only BPE models have pre-tokenizers
if (vocab.type == LLAMA_VOCAB_TYPE_BPE) {
if (tokenizer_pre.empty()) {
JoanFM marked this conversation as resolved.
Show resolved Hide resolved
LLAMA_LOG_WARN("%s: missing pre-tokenizer type, using: 'default'\n", __func__);
LLAMA_LOG_WARN("%s: \n", __func__);
LLAMA_LOG_WARN("%s: ************************************ \n", __func__);
LLAMA_LOG_WARN("%s: GENERATION QUALITY WILL BE DEGRADED! \n", __func__);
LLAMA_LOG_WARN("%s: CONSIDER REGENERATING THE MODEL \n", __func__);
LLAMA_LOG_WARN("%s: ************************************ \n", __func__);
LLAMA_LOG_WARN("%s: \n", __func__);
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
} else if (
if (
tokenizer_pre == "default") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
} else if (
Expand Down Expand Up @@ -4689,7 +4681,8 @@ static void llm_load_vocab(
tokenizer_pre == "jina-es" ||
tokenizer_pre == "jina-de" ||
tokenizer_pre == "jina-v2-es" ||
tokenizer_pre == "jina-v2-de") {
tokenizer_pre == "jina-v2-de" ||
tokenizer_pre == "jina-v2-code") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_GPT2;
} else if (
tokenizer_pre == "refact") {
Expand All @@ -4713,7 +4706,8 @@ static void llm_load_vocab(
tokenizer_pre == "smaug-bpe") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_SMAUG;
} else {
throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
LLAMA_LOG_WARN("%s: missing or unrecognized pre-tokenizer type, using: 'default'\n", __func__);
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
}
} else {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
Expand Down Expand Up @@ -5523,7 +5517,7 @@ static bool llm_load_tensors(

layer.ffn_down_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd});
} else {
layer.ffn_gate = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff});
layer.ffn_gate = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff});
}

layer.layer_out_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd});
Expand Down Expand Up @@ -5564,6 +5558,9 @@ static bool llm_load_tensors(
layer.attn_out_norm = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}); //output_norm
layer.attn_out_norm_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i), {n_embd});

layer.attn_norm_2 = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_NORM_2, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
layer.attn_norm_2_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_NORM_2, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);

layer.ffn_up = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff});
layer.ffn_gate = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff});

Expand Down Expand Up @@ -6626,7 +6623,7 @@ static int llama_model_load(const std::string & fname, llama_model & model, llam
}
} catch (const std::exception & err) {
LLAMA_LOG_ERROR("%s: error loading model: %s\n", __func__, err.what());
return -1;
JoanFM marked this conversation as resolved.
Show resolved Hide resolved
throw;
}

return 0;
Expand Down Expand Up @@ -8527,6 +8524,12 @@ struct llm_build_context {
// attention layer norm
cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_out_norm, model.layers[il].attn_out_norm_b, LLM_NORM, cb, il);

if (model.layers[il].attn_norm_2 != nullptr) {
// re-add the layer input
cur = ggml_add(ctx0, cur, inpL);
cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_norm_2, model.layers[il].attn_norm_2_b, LLM_NORM, cb, il);
}

struct ggml_tensor * ffn_inp = cur;
cb(ffn_inp, "ffn_inp", il);

Expand Down Expand Up @@ -16243,16 +16246,23 @@ struct llama_model * llama_load_model_from_file(
}
model->rpc_servers.push_back(servers);
}
int status = llama_model_load(path_model, *model, params);
GGML_ASSERT(status <= 0);
JoanFM marked this conversation as resolved.
Show resolved Hide resolved
if (status < 0) {
if (status == -1) {
LLAMA_LOG_ERROR("%s: failed to load model\n", __func__);
} else if (status == -2) {
LLAMA_LOG_INFO("%s: cancelled model load\n", __func__);

try {
int status = llama_model_load(path_model, *model, params);
GGML_ASSERT(status <= 0);
if (status < 0) {
if (status == -1) {
LLAMA_LOG_ERROR("%s: failed to load model\n", __func__);
} else if (status == -2) {
LLAMA_LOG_INFO("%s: cancelled model load\n", __func__);
}
delete model;
return nullptr;
}
} catch (...) {
LLAMA_LOG_ERROR("%s: exception loading model\n", __func__);
delete model;
return nullptr;
throw;
}

return model;
Expand Down
Loading