config.yml

# Copyright The Caikit Authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
## Config elements for module boot-time configurations ##

# Disallow downloading from arbitrary URLs in model bootstrapping. This defaults
# to false for security, but can be enabled for bootstrapping
allow_downloads: false
torch_dtype: float32

# Path of folder that will contain all the source prompts
source_prompt_base: ""

# Path for searching base models from
base_models_dir: ""

# Whether or not to purge TGIS prompts on model deletion
unload_tgis_prompt_artifacts: false
# Torchrun elastic launch configuration, e.g., for fine tuning on multiple GPUs
master_addr: localhost
master_port: 29550

training_data_limit:
  __default__: -1
  # Configuration for PeftPromptTuning module
  6655831b-960a-4dc5-8df4-867026e2cd41:
    add_model_name_here: 10000

# Config used only in EmbeddingModule. Set here or use env vars like EMBEDDING_RETRIES=32
embedding:
  # Number of times to retry on error. Most deployments should use 0 retries.
  retries: 0
  # Batch size for encode() if <= 0 or invalid, the sentence-transformers default is used
  batch_size: 0
  # Should implicit truncation (with truncate_input_tokens=0) throw error for truncation (default) or disable this
  implicit_truncation_errors: true
  # Attempt to optimize with PyTorch compile()
  pt2_compile: false
  # Use IPEX optimize. Works best when used with autocast (bfloat16) below.
  ipex: false
  # Use autocast in encode with its default dtype (bfloat16)
  autocast: false
  # For testing, set device to "mps" on MacOS or "xpu" for IPEX GPU.
  # Otherwise, the default does automatic checks for cuda GPU (else cpu).
  device: ""

runtime:
  library: caikit_nlp

# Configure request timeout for TGIS backend (in seconds)
tgis_request_timeout: 60