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

Formatting changes

Browse source
This commit is contained in:
Phillip Kravtsov 2023-10-02 10:26:39 -07:00
parent e6bf87f785
commit cd4d3df820
3 changed files with 9 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml-metal.m
View file

@ -870,8 +870,8 @@ void ggml_metal_graph_compute(
} break; } break;
case GGML_OP_MUL: case GGML_OP_MUL:
{ {
GGML_ASSERT(ggml_is_contiguous(src1));
GGML_ASSERT(ggml_is_contiguous(src0)); GGML_ASSERT(ggml_is_contiguous(src0));
GGML_ASSERT(ggml_is_contiguous(src1));
// utilize float4 // utilize float4
GGML_ASSERT(ne00 % 4 == 0); GGML_ASSERT(ne00 % 4 == 0);
@ -952,9 +952,8 @@ void ggml_metal_graph_compute(
[encoder setComputePipelineState:ctx->pipeline_sqr]; [encoder setComputePipelineState:ctx->pipeline_sqr];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0]; [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1]; [encoder setBuffer:id_dst offset:offs_dst atIndex:1];
//GGML_METAL_LOG_WARN("%s: node %3d, op = %8s dispatching \n", __func__, i, ggml_op_name(dst->op));
const int64_t n = ggml_nelements(dst); const int64_t n = ggml_nelements(dst);
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)]; [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break; } break;

4
gguf-py/gguf/gguf.py
View file

@ -85,7 +85,7 @@ class MODEL_ARCH(IntEnum):
GPTNEOX : int = auto() GPTNEOX : int = auto()
MPT : int = auto() MPT : int = auto()
STARCODER : int = auto() STARCODER : int = auto()
PERSIMMON : int = auto() PERSIMMON : int = auto()
class MODEL_TENSOR(IntEnum): class MODEL_TENSOR(IntEnum):
@ -190,7 +190,7 @@ MODEL_TENSOR_NAMES: dict[MODEL_ARCH, dict[MODEL_TENSOR, str]] = {
MODEL_TENSOR.FFN_UP: "blk.{bid}.ffn_up", MODEL_TENSOR.FFN_UP: "blk.{bid}.ffn_up",
}, },
MODEL_ARCH.PERSIMMON: { MODEL_ARCH.PERSIMMON: {
MODEL_TENSOR.TOKEN_EMBD: "token_embd", MODEL_TENSOR.TOKEN_EMBD: "token_embd",
MODEL_TENSOR.OUTPUT: "output", MODEL_TENSOR.OUTPUT: "output",
MODEL_TENSOR.OUTPUT_NORM: "output_norm", MODEL_TENSOR.OUTPUT_NORM: "output_norm",
MODEL_TENSOR.ATTN_NORM: "blk.{bid}.attn_norm", MODEL_TENSOR.ATTN_NORM: "blk.{bid}.attn_norm",

11
llama.cpp
View file

@ -2473,7 +2473,7 @@ static void llm_load_tensors(
} break; } break;
case LLM_ARCH_PERSIMMON: case LLM_ARCH_PERSIMMON:
{ {
model.tok_embeddings = ml.create_tensor(ctx, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, GGML_BACKEND_CPU); model.tok_embeddings = ml.create_tensor(ctx, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, GGML_BACKEND_CPU);
model.output_norm = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, GGML_BACKEND_CPU); model.output_norm = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, GGML_BACKEND_CPU);
model.output_norm_b = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, GGML_BACKEND_CPU); model.output_norm_b = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, GGML_BACKEND_CPU);
model.output = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, GGML_BACKEND_CPU); model.output = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, GGML_BACKEND_CPU);
@ -2612,8 +2612,8 @@ static bool llama_model_load(
} }
static struct ggml_cgraph * llm_build_llama( static struct ggml_cgraph * llm_build_llama(
llama_context & lctx, llama_context & lctx,
const llama_batch & batch) { const llama_batch & batch) {
const auto & model = lctx.model; const auto & model = lctx.model;
const auto & hparams = model.hparams; const auto & hparams = model.hparams;
const auto & cparams = lctx.cparams; const auto & cparams = lctx.cparams;
@ -4016,9 +4016,8 @@ static struct ggml_cgraph * llm_build_starcoder(
static struct ggml_cgraph * llm_build_persimmon( static struct ggml_cgraph * llm_build_persimmon(
llama_context & lctx, llama_context & lctx,
const llama_batch & batch const llama_batch & batch) {
) {
const auto & model = lctx.model; const auto & model = lctx.model;
const auto & hparams = model.hparams; const auto & hparams = model.hparams;