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

phi3 : duplicate rope factors in each layer #7447

Merged
merged 2 commits into from
May 22, 2024
Merged

phi3 : duplicate rope factors in each layer #7447

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
477973d
phi3 : duplicate rope factors in each layer
slaren May 21, 2024
ef8e9e7
replace bool parameters with named flags
slaren May 22, 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
2 changes: 1 addition & 1 deletion examples/llama-bench/llama-bench.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -195,7 +195,7 @@ static const cmd_params cmd_params_defaults = {
/* model */ {"models/7B/ggml-model-q4_0.gguf"},
/* n_prompt */ {512},
/* n_gen */ {128},
/* n_pg */ {{512, 128}},
/* n_pg */ {},
/* n_batch */ {2048},
/* n_ubatch */ {512},
/* type_k */ {GGML_TYPE_F16},
Expand Down
85 changes: 34 additions & 51 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1940,6 +1940,10 @@ struct llama_layer {
// mamba bias
struct ggml_tensor * ssm_conv1d_b;
struct ggml_tensor * ssm_dt_b;

// long rope factors
struct ggml_tensor * rope_long = nullptr;
struct ggml_tensor * rope_short = nullptr;
};

struct llama_kv_cell {
Expand Down Expand Up @@ -2111,10 +2115,6 @@ struct llama_model {
struct ggml_tensor * output;
struct ggml_tensor * output_b;

// long rope factors
struct ggml_tensor * rope_long = nullptr;
struct ggml_tensor * rope_short = nullptr;

std::vector<llama_layer> layers;

llama_split_mode split_mode;
Expand Down Expand Up @@ -3425,11 +3425,15 @@ struct llama_model_loader {
return get_tensor_meta(get_tensor_name(i));
}

struct ggml_tensor * create_tensor_for(struct ggml_context * ctx, const struct ggml_tensor * cur) {
struct ggml_tensor * create_tensor_for(struct ggml_context * ctx, const struct ggml_tensor * cur, bool duplicated) {
struct ggml_tensor * tensor = ggml_dup_tensor(ctx, cur);
ggml_set_name(tensor, ggml_get_name(cur));

n_created++;
if (duplicated) {
size_data += ggml_nbytes(cur);
} else {
n_created++;
}

return tensor;
}
Expand Down Expand Up @@ -3464,14 +3468,14 @@ struct llama_model_loader {
return cur;
}

struct ggml_tensor * create_tensor(struct ggml_context * ctx, const std::string & name, const std::vector<int64_t> & ne, bool required = true) {
struct ggml_tensor * create_tensor(struct ggml_context * ctx, const std::string & name, const std::vector<int64_t> & ne, bool required = true, bool duplicated = false) {
const struct ggml_tensor * cur = check_tensor_dims(name, ne, required);

if (cur == NULL) {
return NULL;
}

return create_tensor_for(ctx, cur);
return create_tensor_for(ctx, cur, duplicated);
}

struct ggml_tensor * create_tensor_as_view(struct ggml_context * ctx, struct ggml_tensor * base, const std::string & name, const std::vector<int64_t> & ne, size_t offset, bool required = true) {
Expand Down Expand Up @@ -4136,6 +4140,7 @@ static void llm_load_hparams(
switch (hparams.n_layer) {
case 24: model.type = e_model::MODEL_1B; break;
case 32: model.type = e_model::MODEL_3B; break;
case 40: model.type = e_model::MODEL_14B; break;
default: model.type = e_model::MODEL_UNKNOWN;
}
} break;
Expand Down Expand Up @@ -4965,9 +4970,7 @@ static bool llm_load_tensors(
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Was the intention to make the duplicated argument true? I assume that yes, because then the old behavior would be kept.

This seems like required is set to true and leaves duplicated to false.

(this also applies to the other places where model.output is initialized from the token_embd tensor)

Suggested change
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true, true);

Copy link
Member Author

@slaren slaren May 22, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks, I have replaced the boolean parameters with named flags that should make these errors easier to avoid in the future.

}
}
}
Expand Down Expand Up @@ -5045,9 +5048,7 @@ static bool llm_load_tensors(
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
}
}

Expand Down Expand Up @@ -5174,9 +5175,7 @@ static bool llm_load_tensors(

model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
if (!model.output) {
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}); // needs to be on GPU
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true); // needs to be on GPU
}
}

Expand Down Expand Up @@ -5211,9 +5210,7 @@ static bool llm_load_tensors(
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
if (!model.output) {
// needs to be on GPU
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
}

}
Expand Down Expand Up @@ -5389,9 +5386,7 @@ static bool llm_load_tensors(

model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
if (!model.output) {
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}); // needs to be on GPU
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true); // needs to be on GPU
}
}

Expand Down Expand Up @@ -5511,9 +5506,7 @@ static bool llm_load_tensors(
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
}
}

Expand Down Expand Up @@ -5639,9 +5632,6 @@ static bool llm_load_tensors(
{
model.tok_embd = ml.create_tensor(ctx_input, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab });

model.rope_long = ml.create_tensor(ctx_input, tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight"), { n_embd_head/2 }, false);
model.rope_short = ml.create_tensor(ctx_input, tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight"), { n_embd_head/2 }, false);

// output
{
model.output_norm = ml.create_tensor(ctx_output, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd });
Expand All @@ -5663,6 +5653,9 @@ static bool llm_load_tensors(

layer.ffn_down = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd });
layer.ffn_up = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP, "weight", i), { n_embd, 2 * n_ff });

layer.rope_long = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight"), { n_embd_head/2 }, false, i != 0);
layer.rope_short = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight"), { n_embd_head/2 }, false, i != 0);
}
} break;
case LLM_ARCH_PLAMO:
Expand Down Expand Up @@ -5831,9 +5824,7 @@ static bool llm_load_tensors(

// output
model.output_norm = ml.create_tensor(ctx_output, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}); // same as tok_embd, duplicated to allow offloading
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true); // same as tok_embd, duplicated to allow offloading

const int64_t n_ff = hparams.n_ff;
const int64_t n_embd_head_k = hparams.n_embd_head_k;
Expand Down Expand Up @@ -5871,9 +5862,7 @@ static bool llm_load_tensors(
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
}

}
Expand Down Expand Up @@ -5927,9 +5916,7 @@ static bool llm_load_tensors(
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
// if output is NULL, init from the input tok embed, duplicated to allow offloading
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
}
}

Expand Down Expand Up @@ -5990,9 +5977,7 @@ static bool llm_load_tensors(
{
model.output_norm = ml.create_tensor(ctx_output, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
// init output from the input tok embed
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
}

for (int i = 0; i < n_layer; ++i) {
Expand Down Expand Up @@ -6027,9 +6012,7 @@ static bool llm_load_tensors(
model.output = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, false);
// if output is NULL, init from the input tok embed
if (model.output == NULL) {
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab});
ml.n_created--; // artificial tensor
ml.size_data += ggml_nbytes(model.output);
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, true);
}
}

Expand Down Expand Up @@ -6872,9 +6855,9 @@ struct llm_build_context {
cb(lctx.inp_K_shift, "K_shift", -1);
ggml_set_input(lctx.inp_K_shift);

struct ggml_tensor * rope_factors = build_rope_factors();

for (int il = 0; il < n_layer; ++il) {
struct ggml_tensor * rope_factors = build_rope_factors(il);
struct ggml_tensor * tmp =
// we rotate only the first n_rot dimensions
ggml_rope_ext_inplace(ctx0,
Expand Down Expand Up @@ -6988,15 +6971,15 @@ struct llm_build_context {
return lctx.inp_pos;
}

struct ggml_tensor * build_rope_factors() {
struct ggml_tensor * build_rope_factors(int il) {
// choose long/short freq factors based on the context size
const auto n_ctx_pre_seq = cparams.n_ctx / cparams.n_seq_max;

if (n_ctx_pre_seq > hparams.n_yarn_orig_ctx) {
return model.rope_long;
return model.layers[il].rope_long;
}

return model.rope_short;
return model.layers[il].rope_short;
}

struct ggml_tensor * build_inp_out_ids() {
Expand Down Expand Up @@ -9117,14 +9100,14 @@ struct llm_build_context {
// KQ_mask (mask for 1 head, it will be broadcasted to all heads)
struct ggml_tensor * KQ_mask = build_inp_KQ_mask();

// rope freq factors for 128k context
struct ggml_tensor * rope_factors = build_rope_factors();

for (int il = 0; il < n_layer; ++il) {
auto residual = inpL;

// self-attention
{
// rope freq factors for 128k context
struct ggml_tensor * rope_factors = build_rope_factors(il);

struct ggml_tensor* attn_norm_output = llm_build_norm(ctx0, inpL, hparams,
model.layers[il].attn_norm,
NULL,
Expand Down
Loading