fix: Allow using unsupported base models without adapter loading (#76)

launcher/src/main.rs

@@ -74,10 +74,10 @@ impl std::fmt::Display for Dtype {
 struct Args {
     /// The name of the model to load.
     /// Can be a MODEL_ID as listed on <https://hf.co/models> like
-    /// `gpt2` or `OpenAssistant/oasst-sft-1-pythia-12b`.
+    /// `gpt2` or `mistralai/Mistral-7B-Instruct-v0.1`.
     /// Or it can be a local directory containing the necessary files
     /// as saved by `save_pretrained(...)` methods of transformers
-    #[clap(default_value = "bigscience/bloom-560m", long, env)]
+    #[clap(default_value = "mistralai/Mistral-7B-Instruct-v0.1", long, env)]
     model_id: String,
 
     /// The name of the adapter to load.

server/lorax_server/models/model.py

@@ -3,10 +3,11 @@
 
 from abc import ABC, abstractmethod
 from typing import List, Tuple, Optional, TypeVar, Type
-from transformers import PreTrainedTokenizerBase, PretrainedConfig
+from transformers import PreTrainedTokenizerBase
 
 from lorax_server.models.types import Batch, GeneratedText
 from lorax_server.pb.generate_pb2 import InfoResponse
+from lorax_server.utils.adapter import BASE_MODEL_ADAPTER_ID
 
 B = TypeVar("B", bound=Batch)
 
@@ -101,3 +102,8 @@ def check_initialized(self):
             raise RuntimeError(
                 f"found uninitialized parameters in model {self.__class__.__name__}: {uninitialized_parameters}"
             )
+
+    def load_adapter(self, adapter_id, adapter_source, adapter_index):
+        if adapter_id == BASE_MODEL_ADAPTER_ID:
+            return
+        raise ValueError("This model does not support adapter loading.")