This commit is contained in:
OlivierDehaene 2023-06-01 19:49:13 +02:00
parent 95d3546976
commit e7248fe90e
6 changed files with 17 additions and 11 deletions

8
Cargo.lock generated
View File

@ -2557,7 +2557,7 @@ dependencies = [
[[package]] [[package]]
name = "text-generation-benchmark" name = "text-generation-benchmark"
version = "0.8.1" version = "0.8.2"
dependencies = [ dependencies = [
"average", "average",
"clap", "clap",
@ -2577,7 +2577,7 @@ dependencies = [
[[package]] [[package]]
name = "text-generation-client" name = "text-generation-client"
version = "0.8.1" version = "0.8.2"
dependencies = [ dependencies = [
"futures", "futures",
"grpc-metadata", "grpc-metadata",
@ -2593,7 +2593,7 @@ dependencies = [
[[package]] [[package]]
name = "text-generation-launcher" name = "text-generation-launcher"
version = "0.8.1" version = "0.8.2"
dependencies = [ dependencies = [
"clap", "clap",
"ctrlc", "ctrlc",
@ -2609,7 +2609,7 @@ dependencies = [
[[package]] [[package]]
name = "text-generation-router" name = "text-generation-router"
version = "0.8.1" version = "0.8.2"
dependencies = [ dependencies = [
"async-stream", "async-stream",
"axum", "axum",

View File

@ -8,7 +8,7 @@ members = [
] ]
[workspace.package] [workspace.package]
version = "0.8.1" version = "0.8.2"
edition = "2021" edition = "2021"
authors = ["Olivier Dehaene"] authors = ["Olivier Dehaene"]
homepage = "https://github.com/huggingface/text-generation-inference" homepage = "https://github.com/huggingface/text-generation-inference"

View File

@ -10,7 +10,7 @@
"name": "Apache 2.0", "name": "Apache 2.0",
"url": "https://www.apache.org/licenses/LICENSE-2.0" "url": "https://www.apache.org/licenses/LICENSE-2.0"
}, },
"version": "0.8.1" "version": "0.8.2"
}, },
"paths": { "paths": {
"/": { "/": {

View File

@ -1,6 +1,6 @@
[tool.poetry] [tool.poetry]
name = "text-generation-server" name = "text-generation-server"
version = "0.8.1" version = "0.8.2"
description = "Text Generation Inference Python gRPC Server" description = "Text Generation Inference Python gRPC Server"
authors = ["Olivier Dehaene <olivier@huggingface.co>"] authors = ["Olivier Dehaene <olivier@huggingface.co>"]

View File

@ -138,7 +138,9 @@ def get_model(
trust_remote_code=trust_remote_code, trust_remote_code=trust_remote_code,
) )
config_dict, _ = PretrainedConfig.get_config_dict(model_id, revision=revision, trust_remote_code=trust_remote_code) config_dict, _ = PretrainedConfig.get_config_dict(
model_id, revision=revision, trust_remote_code=trust_remote_code
)
model_type = config_dict["model_type"] model_type = config_dict["model_type"]
if model_type == "gpt_bigcode": if model_type == "gpt_bigcode":

View File

@ -87,7 +87,9 @@ class FlashSantacoder(FlashCausalLM):
): ):
for filename in filenames: for filename in filenames:
with safe_open( with safe_open(
filename, framework="pt", device=str(device) if quantize is None else "cpu" filename,
framework="pt",
device=str(device) if quantize is None else "cpu",
) as f: ) as f:
for key in f.keys(): for key in f.keys():
value = f.get_tensor(key) value = f.get_tensor(key)
@ -148,11 +150,13 @@ class FlashSantacoder(FlashCausalLM):
module._parameters[param_name][: value.shape[0]] = value module._parameters[param_name][: value.shape[0]] = value
elif "kv_attn.weight" in key: elif "kv_attn.weight" in key:
module._parameters[param_name][ module._parameters[param_name][
model.transformer.head_size * model.transformer.num_heads : model.transformer.head_size
* model.transformer.num_heads :
] = value ] = value
elif "kv_attn.bias" in key: elif "kv_attn.bias" in key:
module._parameters[param_name][ module._parameters[param_name][
model.transformer.head_size * model.transformer.num_heads : model.transformer.head_size
* model.transformer.num_heads :
] = value ] = value
else: else:
if current_parameter_tensor.shape != value.shape: if current_parameter_tensor.shape != value.shape: