Shark-1.0: Update gradio to 4.19.2 (#2098)

* Shark-1.0: Update gradio to 4.19.2 - Move pin for gradio from 4.12 -> 4.19.2 on SHARK-1.0 branch - Fix png import txt2img/txt2img_sdxl event handling for gradio 4.19.2 - Fix sendto txt2img/txt2img_sdxl event handling for gradio 4.19.2 * Shark-1.0: Reformat with black - Apply the new (2024) black reformattings
nod-ai · Mar 1, 2024 · 9fc0b5b · 9fc0b5b
1 parent 85ae2ee
commit 9fc0b5b
Show file tree

Hide file tree

Showing 31 changed files with 281 additions and 225 deletions.
diff --git a/.gitignore b/.gitignore
@@ -168,6 +168,7 @@ cython_debug/
 *venv/
 shark_tmp/
 *.vmfb
+*.safetensors
 .use-iree
 tank/dict_configs.py
 *.csv

diff --git a/apps/language_models/langchain/cli.py b/apps/language_models/langchain/cli.py
@@ -151,9 +151,9 @@ def run_cli(  # for local function:
             eval_vars[eval_func_param_names.index("iinput")] = eval_vars[
                 eval_func_param_names.index("iinput_nochat")
             ] = ""  # no input yet
-            eval_vars[
-                eval_func_param_names.index("context")
-            ] = ""  # no context yet
+            eval_vars[eval_func_param_names.index("context")] = (
+                ""  # no context yet
+            )
 
             # grab other parameters, like langchain_mode
             for k in eval_func_param_names:

diff --git a/apps/language_models/langchain/create_data.py b/apps/language_models/langchain/create_data.py
@@ -336,14 +336,16 @@ def test_config_to_json():
                             "\n", ""
                         ),
                     },
-                    {
-                        "prompt_type": "plain",
-                        "instruction": f"<human>: How can I do this: {title}.\n<bot>: Set the {k.replace('_', ' ')} config.toml\n<human>:".replace(
-                            "\n", ""
-                        ),
-                    }
-                    if title and comment
-                    else None,
+                    (
+                        {
+                            "prompt_type": "plain",
+                            "instruction": f"<human>: How can I do this: {title}.\n<bot>: Set the {k.replace('_', ' ')} config.toml\n<human>:".replace(
+                                "\n", ""
+                            ),
+                        }
+                        if title and comment
+                        else None
+                    ),
                     {
                         "prompt_type": "human_bot",
                         "instruction": f"Explain the following expert setting for Driverless AI",

diff --git a/apps/language_models/langchain/gen.py b/apps/language_models/langchain/gen.py
@@ -582,19 +582,19 @@ def get_hf_model(
                 tokenizer,
                 model=base_model,
                 device=0 if self.device == "cuda" else -1,
-                torch_dtype=torch.float16
-                if self.device == "cuda"
-                else torch.float32,
+                torch_dtype=(
+                    torch.float16 if self.device == "cuda" else torch.float32
+                ),
             )
         else:
             assert self.device in ["cuda", "cpu", "mps"], (
                 "Unsupported device %s" % self.device
             )
             model_kwargs = dict(
                 local_files_only=local_files_only,
-                torch_dtype=torch.float16
-                if self.device == "cuda"
-                else torch.float32,
+                torch_dtype=(
+                    torch.float16 if self.device == "cuda" else torch.float32
+                ),
                 resume_download=resume_download,
                 use_auth_token=use_auth_token,
                 trust_remote_code=trust_remote_code,
@@ -628,9 +628,9 @@ def get_hf_model(
                 # MPT doesn't support spreading over GPUs
                 model_kwargs.update(
                     dict(
-                        device_map={"": gpu_id}
-                        if self.device == "cuda"
-                        else "cpu"
+                        device_map=(
+                            {"": gpu_id} if self.device == "cuda" else "cpu"
+                        )
                     )
                 )
 
@@ -689,17 +689,19 @@ def get_hf_model(
                 model = PeftModel.from_pretrained(
                     model,
                     lora_weights,
-                    torch_dtype=torch.float16
-                    if self.device == "cuda"
-                    else torch.float32,
+                    torch_dtype=(
+                        torch.float16
+                        if self.device == "cuda"
+                        else torch.float32
+                    ),
                     local_files_only=local_files_only,
                     resume_download=resume_download,
                     use_auth_token=use_auth_token,
                     trust_remote_code=trust_remote_code,
                     offload_folder=offload_folder,
-                    device_map={"": 0}
-                    if self.device == "cuda"
-                    else {"": "cpu"},  # seems to be required
+                    device_map=(
+                        {"": 0} if self.device == "cuda" else {"": "cpu"}
+                    ),  # seems to be required
                 )
             else:
                 with torch.device(self.device):
@@ -716,9 +718,11 @@ def get_hf_model(
                     model = PeftModel.from_pretrained(
                         model,
                         lora_weights,
-                        torch_dtype=torch.float16
-                        if self.device == "cuda"
-                        else torch.float32,
+                        torch_dtype=(
+                            torch.float16
+                            if self.device == "cuda"
+                            else torch.float32
+                        ),
                         local_files_only=local_files_only,
                         resume_download=resume_download,
                         use_auth_token=use_auth_token,
@@ -1143,9 +1147,9 @@ def evaluate(
                 db=db1,
                 user_path=user_path,
                 detect_user_path_changes_every_query=detect_user_path_changes_every_query,
-                cut_distanct=1.1
-                if langchain_mode in ["wiki_full"]
-                else 1.64,  # FIXME, too arbitrary
+                cut_distanct=(
+                    1.1 if langchain_mode in ["wiki_full"] else 1.64
+                ),  # FIXME, too arbitrary
                 use_openai_embedding=use_openai_embedding,
                 use_openai_model=use_openai_model,
                 hf_embedding_model=hf_embedding_model,
@@ -1523,9 +1527,11 @@ def mean(a):""",
         examples += [
             [
                 summarize_example1,
-                "Summarize"
-                if prompt_type not in ["plain", "instruct_simple"]
-                else "",
+                (
+                    "Summarize"
+                    if prompt_type not in ["plain", "instruct_simple"]
+                    else ""
+                ),
             ]
             + params_list
         ]
@@ -1548,14 +1554,14 @@ def mean(a):""",
             ]
             # adjust examples if non-chat mode
             if not chat:
-                example[
-                    eval_func_param_names.index("instruction_nochat")
-                ] = example[eval_func_param_names.index("instruction")]
+                example[eval_func_param_names.index("instruction_nochat")] = (
+                    example[eval_func_param_names.index("instruction")]
+                )
                 example[eval_func_param_names.index("instruction")] = ""
 
-                example[
-                    eval_func_param_names.index("iinput_nochat")
-                ] = example[eval_func_param_names.index("iinput")]
+                example[eval_func_param_names.index("iinput_nochat")] = (
+                    example[eval_func_param_names.index("iinput")]
+                )
                 example[eval_func_param_names.index("iinput")] = ""
             assert len(example) == len(
                 eval_func_param_names

diff --git a/apps/language_models/langchain/gpt_langchain.py b/apps/language_models/langchain/gpt_langchain.py
@@ -489,9 +489,9 @@ def _call(
         chunk = True
         chunk_size = 512
         client_kwargs = dict(
-            instruction=prompt
-            if self.chat_client
-            else "",  # only for chat=True
+            instruction=(
+                prompt if self.chat_client else ""
+            ),  # only for chat=True
             iinput="",  # only for chat=True
             context="",
             # streaming output is supported, loops over and outputs each generation in streaming mode

diff --git a/apps/language_models/langchain/h2oai_pipeline.py b/apps/language_models/langchain/h2oai_pipeline.py
@@ -316,9 +316,9 @@ def generate_token(h2ogpt_shark_model, model, tokenizer, **generate_kwargs):
     )
 
     model_kwargs["output_attentions"] = generation_config.output_attentions
-    model_kwargs[
-        "output_hidden_states"
-    ] = generation_config.output_hidden_states
+    model_kwargs["output_hidden_states"] = (
+        generation_config.output_hidden_states
+    )
     model_kwargs["use_cache"] = generation_config.use_cache
 
     input_ids = (

diff --git a/apps/language_models/langchain/image_captions.py b/apps/language_models/langchain/image_captions.py
@@ -7,6 +7,7 @@
 https://huggingface.co/Salesforce/blip-image-captioning-base
 
 """
+
 from typing import List, Union, Any, Tuple
 
 import requests

diff --git a/apps/language_models/langchain/read_wiki_full.py b/apps/language_models/langchain/read_wiki_full.py
@@ -1,4 +1,5 @@
 """Load Data from a MediaWiki dump xml."""
+
 import ast
 import glob
 import pickle
@@ -105,9 +106,11 @@ def load(self) -> List[Document]:
                     source="https://en.wikipedia.org/wiki/" + title_url,
                     id=page.id,
                     redirect=page.redirect,
-                    views=self.views[page.title]
-                    if self.views is not None
-                    else -1,
+                    views=(
+                        self.views[page.title]
+                        if self.views is not None
+                        else -1
+                    ),
                 )
                 metadata = {k: v for k, v in metadata.items() if v is not None}
                 docs.append(Document(page_content=text, metadata=metadata))

diff --git a/apps/language_models/src/pipelines/falcon_pipeline.py b/apps/language_models/src/pipelines/falcon_pipeline.py
@@ -279,16 +279,18 @@ def compile_layer(
         path = shark_module.save_module(
             self.falcon_vmfb_path.parent.absolute(),
             self.falcon_vmfb_path.stem,
-            extra_args=[
-                "--iree-vm-target-truncate-unsupported-floats",
-                "--iree-codegen-check-ir-before-llvm-conversion=false",
-                "--iree-vm-bytecode-module-output-format=flatbuffer-binary",
-            ]
-            + [
-                "--iree-llvmcpu-use-fast-min-max-ops",
-            ]
-            if self.precision == "int4"
-            else [],
+            extra_args=(
+                [
+                    "--iree-vm-target-truncate-unsupported-floats",
+                    "--iree-codegen-check-ir-before-llvm-conversion=false",
+                    "--iree-vm-bytecode-module-output-format=flatbuffer-binary",
+                ]
+                + [
+                    "--iree-llvmcpu-use-fast-min-max-ops",
+                ]
+                if self.precision == "int4"
+                else []
+            ),
             debug=self.debug,
         )
         print("Saved falcon vmfb at ", str(path))
@@ -326,9 +328,11 @@ def compile(self):
             lm_head,
             [sample_hidden_states],
             "lm_head",
-            device_idx=(0 % num_devices) % args.num_shards
-            if self.device == "rocm"
-            else None,
+            device_idx=(
+                (0 % num_devices) % args.num_shards
+                if self.device == "rocm"
+                else None
+            ),
         )
         shark_lm_head = CompiledLMHeadEmbeddingLayer(shark_lm_head)
 
@@ -340,9 +344,11 @@ def compile(self):
             word_embedding,
             [sample_input_ids],
             "word_embeddings",
-            device_idx=(1 % num_devices) % args.num_shards
-            if self.device == "rocm"
-            else None,
+            device_idx=(
+                (1 % num_devices) % args.num_shards
+                if self.device == "rocm"
+                else None
+            ),
         )
         shark_word_embedding = CompiledWordEmbeddingsLayer(
             shark_word_embedding
@@ -354,9 +360,11 @@ def compile(self):
             ln_f,
             [sample_hidden_states],
             "ln_f",
-            device_idx=(2 % num_devices) % args.num_shards
-            if self.device == "rocm"
-            else None,
+            device_idx=(
+                (2 % num_devices) % args.num_shards
+                if self.device == "rocm"
+                else None
+            ),
         )
         shark_ln_f = CompiledLNFEmbeddingLayer(shark_ln_f)
 
@@ -458,9 +466,9 @@ def generate(self, prompt):
         )
 
         model_kwargs["output_attentions"] = generation_config.output_attentions
-        model_kwargs[
-            "output_hidden_states"
-        ] = generation_config.output_hidden_states
+        model_kwargs["output_hidden_states"] = (
+            generation_config.output_hidden_states
+        )
         model_kwargs["use_cache"] = generation_config.use_cache
 
         input_ids = (
@@ -790,16 +798,18 @@ def compile(self):
         path = shark_module.save_module(
             self.falcon_vmfb_path.parent.absolute(),
             self.falcon_vmfb_path.stem,
-            extra_args=[
-                "--iree-vm-target-truncate-unsupported-floats",
-                "--iree-codegen-check-ir-before-llvm-conversion=false",
-                "--iree-vm-bytecode-module-output-format=flatbuffer-binary",
-            ]
-            + [
-                "--iree-llvmcpu-use-fast-min-max-ops",
-            ]
-            if self.precision == "int4"
-            else [],
+            extra_args=(
+                [
+                    "--iree-vm-target-truncate-unsupported-floats",
+                    "--iree-codegen-check-ir-before-llvm-conversion=false",
+                    "--iree-vm-bytecode-module-output-format=flatbuffer-binary",
+                ]
+                + [
+                    "--iree-llvmcpu-use-fast-min-max-ops",
+                ]
+                if self.precision == "int4"
+                else []
+            ),
             debug=self.debug,
         )
         print("Saved falcon vmfb at ", str(path))
@@ -859,9 +869,9 @@ def generate(self, prompt):
         batch_size = inputs_tensor.shape[0]
 
         model_kwargs["output_attentions"] = generation_config.output_attentions
-        model_kwargs[
-            "output_hidden_states"
-        ] = generation_config.output_hidden_states
+        model_kwargs["output_hidden_states"] = (
+            generation_config.output_hidden_states
+        )
         model_kwargs["use_cache"] = generation_config.use_cache
 
         input_ids = (

diff --git a/apps/language_models/src/pipelines/minigpt4_utils/blip_processors.py b/apps/language_models/src/pipelines/minigpt4_utils/blip_processors.py
@@ -4,6 +4,7 @@
  SPDX-License-Identifier: BSD-3-Clause
  For full license text, see the LICENSE_Lavis file in the repo root or https://opensource.org/licenses/BSD-3-Clause
 """
+
 from omegaconf import OmegaConf
 from torchvision import transforms
 from torchvision.transforms.functional import InterpolationMode

diff --git a/apps/language_models/src/pipelines/minigpt4_utils/eva_vit.py b/apps/language_models/src/pipelines/minigpt4_utils/eva_vit.py
@@ -383,9 +383,9 @@ def __init__(
         super().__init__()
         self.image_size = img_size
         self.num_classes = num_classes
-        self.num_features = (
-            self.embed_dim
-        ) = embed_dim  # num_features for consistency with other models
+        self.num_features = self.embed_dim = (
+            embed_dim  # num_features for consistency with other models
+        )
 
         self.patch_embed = PatchEmbed(
             img_size=img_size,
@@ -429,9 +429,11 @@ def __init__(
                     drop_path=dpr[i],
                     norm_layer=norm_layer,
                     init_values=init_values,
-                    window_size=self.patch_embed.patch_shape
-                    if use_rel_pos_bias
-                    else None,
+                    window_size=(
+                        self.patch_embed.patch_shape
+                        if use_rel_pos_bias
+                        else None
+                    ),
                 )
                 for i in range(depth)
             ]