Merge from GO: fixes for --help

automation/script/help.py

@@ -32,20 +32,26 @@ def display_help(self_module, input_params):
 
     scripts_list = search_result['list']
     if not scripts_list:
-        return {'return': 1, 'error': 'No scripts were found'}
 
-    generic_inputs = self_module.input_flags_converted_to_env
+        print("")
+        print("Please use script tags or alias/uid to get help for a specific script")
+        print("")
+        print("Generic Inputs for all Scripts:")
+        print("")
+        print_input_descriptions(generic_inputs)
 
-    # Step 4: Iterate over scripts and generate help output
-    for script in sorted(scripts_list, key=lambda x: x.meta.get('alias', '')):
-        metadata = script.meta
-        script_path = script.path
-        print_script_help(
-            metadata,
-            script_path,
-            generic_inputs,
-            env,
-            self_module)
+    else:
+        # Step 4: Iterate over scripts and generate help output
+        for script in sorted(
+                scripts_list, key=lambda x: x.meta.get('alias', '')):
+            metadata = script.meta
+            script_path = script.path
+            print_script_help(
+                metadata,
+                script_path,
+                generic_inputs,
+                env,
+                self_module)
 
     return {'return': 0}
 

automation/script/module.py

@@ -2870,6 +2870,8 @@ def search(self, i):
             if i.get(key):
                 ii[key] = i[key]
 
+        if i.get('all'):
+            ii['all'] = i['all']
         r = super(ScriptAutomation, self).search(ii)
         if r['return'] > 0:
             return r

script/app-mlperf-inference-mlcommons-python/README.md

@@ -1,10 +1,10 @@
 # README for app-mlperf-inference-mlcommons-python
 This README is automatically generated. Add custom content in [info.md](info.md). Please follow the [script execution document](https://docs.mlcommons.org/mlcflow/targets/script/execution-flow/) to understand more about the MLC script execution.
 
-`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/user`, you can do
+`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/$USER`, you can do
 ```
-mkdir /mnt/user/MLC
-ln -s /mnt/user/MLC $HOME/MLC
+mkdir /mnt/$USER/MLC
+ln -s /mnt/$USER/MLC $HOME/MLC
 ```
 You can also use the `ENV` variable `MLC_REPOS` to control this location but this will need a set after every system reboot.
 
@@ -136,7 +136,7 @@ mlcr app,vision,language,mlcommons,mlperf,inference,reference,ref
 - `llama2-70b-99` (base: llama2-70b_)
 - `llama2-70b-99.9` (base: llama2-70b_)
 - `llama3_1-405b`
-- `llama3_1-8b`
+- `llama3_1-8b_`
 - `mixtral-8x7b`
 - `pointpainting`
 - `resnet50` (default)
@@ -166,6 +166,8 @@ mlcr app,vision,language,mlcommons,mlperf,inference,reference,ref
 - `dlrm-v2_`
 - `gptj_`
 - `llama2-70b_`
+- `llama3_1-8b` (base: llama3_1-8b_)
+- `llama3_1-8b-edge` (base: llama3_1-8b_)
 - `multistream`
 - `offline`
 - `r2.1_default`

script/app-mlperf-inference-nvidia/README.md

@@ -1,13 +1,96 @@
 # README for app-mlperf-inference-nvidia
-This README is automatically generated. Add custom content in [info.txt](info.txt). Please follow the [script execution document](https://docs.mlcommons.org/mlcflow/targets/script/execution-flow/) to understand more about the MLC script execution.
+This README is automatically generated. Add custom content in [info.md](info.md). Please follow the [script execution document](https://docs.mlcommons.org/mlcflow/targets/script/execution-flow/) to understand more about the MLC script execution.
 
+`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/$USER`, you can do
+```
+mkdir /mnt/$USER/MLC
+ln -s /mnt/$USER/MLC $HOME/MLC
+```
+You can also use the `ENV` variable `MLC_REPOS` to control this location but this will need a set after every system reboot.
+
+## Setup
+
+If you are not on a Python development environment please refer to the [official docs](https://docs.mlcommons.org/mlcflow/install/) for the installation.
+
+```bash
+python3 -m venv mlcflow
+. mlcflow/bin/activate
+pip install mlcflow
+```
+
+- Using a virtual environment is recommended (per `pip` best practices), but you may skip it or use `--break-system-packages` if needed.
+
+### Pull mlperf-automations
+
+Once `mlcflow` is installed:
+
+```bash
+mlc pull repo mlcommons@mlperf-automations --pat=<Your Private Access Token>
+```
+- `--pat` or `--ssh` is only needed if the repo is PRIVATE
+- If `--pat` is avoided, you'll be asked to enter the password where you can enter your Private Access Token
+- `--ssh` option can be used instead of `--pat=<>` option if you prefer to use SSH for accessing the github repository.
 ## Run Commands
 
 ```bash
 mlcr reproduce,mlcommons,mlperf,inference,harness,nvidia-harness,nvidia
 ```
 
-No script specific inputs
+### Script Inputs
+
+| Name | Description | Choices | Default |
+|------|-------------|---------|------|
+| `--count` |  |  | `` |
+| `--max_batchsize` |  |  | `` |
+| `--mlperf_conf` |  |  | `` |
+| `--mode` |  |  | `performance` |
+| `--output_dir` |  |  | `` |
+| `--scenario` |  |  | `Offline` |
+| `--user_conf` |  |  | `` |
+| `--devices` |  |  | `` |
+| `--skip_preprocess` |  |  | `no` |
+| `--skip_preprocessing` | Alias for skip_preprocess |  | `` |
+| `--target_qps` |  |  | `` |
+| `--offline_target_qps` |  |  | `` |
+| `--server_target_qps` |  |  | `` |
+| `--target_latency` |  |  | `` |
+| `--singlestream_target_latency` |  |  | `` |
+| `--multistream_target_latency` |  |  | `` |
+| `--use_triton` |  |  | `` |
+| `--gpu_copy_streams` |  |  | `` |
+| `--gpu_inference_streams` |  |  | `` |
+| `--gpu_batch_size` |  |  | `` |
+| `--dla_copy_streams` |  |  | `` |
+| `--dla_inference_streams` |  |  | `` |
+| `--dla_batch_size` |  |  | `` |
+| `--input_format` |  |  | `` |
+| `--performance_sample_count` |  |  | `` |
+| `--workspace_size` |  |  | `` |
+| `--log_dir` |  |  | `` |
+| `--use_graphs` |  |  | `` |
+| `--run_infer_on_copy_streams` |  |  | `` |
+| `--start_from_device` |  |  | `` |
+| `--end_on_device` |  |  | `` |
+| `--max_dlas` |  |  | `` |
+| `--power_setting` |  |  | `` |
+| `--make_cmd` |  |  | `` |
+| `--rerun` |  |  | `` |
+| `--extra_run_options` |  |  | `` |
+| `--use_deque_limit` |  |  | `` |
+| `--deque_timeout_usec` |  |  | `` |
+| `--use_cuda_thread_per_device` |  |  | `` |
+| `--num_warmups` |  |  | `` |
+| `--graphs_max_seqlen` |  |  | `` |
+| `--num_issue_query_threads` |  |  | `` |
+| `--soft_drop` |  |  | `` |
+| `--use_small_tile_gemm_plugin` |  |  | `` |
+| `--audio_buffer_num_lines` |  |  | `` |
+| `--use_fp8` |  |  | `` |
+| `--enable_sort` |  |  | `` |
+| `--num_sort_segments` |  |  | `` |
+| `--skip_postprocess` |  |  | `` |
+| `--embedding_weights_on_gpu_part` |  |  | `` |
+| `--sdxl_batcher_time_limit` |  |  | `` |
 ### Generic Script Inputs
 
 | Name | Description | Choices | Default |
@@ -26,3 +109,128 @@ No script specific inputs
 | `--gh_token` | Github Token |  | `` |
 | `--hf_token` | Huggingface Token |  | `` |
 | `--verify_ssl` | Verify SSL |  | `False` |
+## Variations
+
+### Backend
+
+- `tensorrt` (default)
+
+### Batch-size
+
+- `batch_size.#` _(# can be substituted dynamically)_
+
+### Batchsize-format-change
+
+- `pre5.0`
+- `v5.0+`
+
+### Build-engine-options
+
+- `build_engine_options.#` _(# can be substituted dynamically)_
+
+### Device
+
+- `cpu`
+- `cuda` (default)
+
+### Device-memory
+
+- `gpu_memory.#` _(# can be substituted dynamically)_
+- `gpu_memory.16`
+- `gpu_memory.24`
+- `gpu_memory.32`
+- `gpu_memory.40`
+- `gpu_memory.48`
+- `gpu_memory.8`
+- `gpu_memory.80`
+
+### Dla-batch-size
+
+- `dla_batch_size.#` _(# can be substituted dynamically)_
+
+### Gpu-connection
+
+- `pcie`
+- `sxm`
+
+### Gpu-name
+
+- `a100`
+- `a6000`
+- `custom`
+- `l4`
+- `orin`
+- `rtx_4090`
+- `rtx_6000_ada`
+- `t4`
+
+### Graphs
+
+- `use-graphs`
+
+### Loadgen-scenario
+
+- `multistream`
+- `offline`
+- `server`
+- `singlestream`
+
+### Model
+
+- `3d-unet-99` (base: 3d-unet_)
+- `3d-unet-99.9` (base: 3d-unet_)
+- `bert-99` (base: bert_)
+- `bert-99.9` (base: bert_)
+- `dlrm-v2-99` (base: dlrm_)
+- `dlrm-v2-99.9` (base: dlrm_)
+- `gptj-99` (base: gptj_)
+- `gptj-99.9` (base: gptj_)
+- `llama2-70b-99` (base: llama2-70b_)
+- `llama2-70b-99.9` (base: llama2-70b_)
+- `resnet50` (default)
+- `retinanet`
+- `rnnt`
+- `sdxl`
+
+### Num-gpus
+
+- `num-gpus.#` _(# can be substituted dynamically)_
+- `num-gpus.1` (default)
+
+### Power-mode
+
+- `maxn`
+- `maxq`
+
+### Run-mode
+
+- `build`
+- `build_engine` (alias: build-engine)
+- `calibrate`
+- `download_model`
+- `prebuild`
+- `preprocess_data` (alias: preprocess-data)
+- `run_harness` (default)
+
+### Triton
+
+- `use_triton`
+
+### Ungrouped
+
+- `3d-unet_`
+- `bert_`
+- `default_variations`
+- `dlrm_`
+- `env`
+- `gptj_`
+- `llama2-70b_`
+- `run-harness`
+- `v3.1` (base: pre5.0)
+
+### Version
+
+- `v4.0` (base: pre5.0)
+- `v4.1` (base: pre5.0)
+- `v4.1-dev` (base: pre5.0) (default)
+- `v5.0` (base: v5.0+)

script/app-mlperf-inference/README.md

@@ -1,10 +1,10 @@
 # README for app-mlperf-inference
 This README is automatically generated. Add custom content in [info.md](info.md). Please follow the [script execution document](https://docs.mlcommons.org/mlcflow/targets/script/execution-flow/) to understand more about the MLC script execution.
 
-`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/user`, you can do
+`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/$USER`, you can do
 ```
-mkdir /mnt/user/MLC
-ln -s /mnt/user/MLC $HOME/MLC
+mkdir /mnt/$USER/MLC
+ln -s /mnt/$USER/MLC $HOME/MLC
 ```
 You can also use the `ENV` variable `MLC_REPOS` to control this location but this will need a set after every system reboot.
 
@@ -166,7 +166,7 @@ mlcr app,vision,language,mlcommons,mlperf,inference,generic
 - `llama2-70b-99` (base: llama2-70b_)
 - `llama2-70b-99.9` (base: llama2-70b_)
 - `llama3_1-405b`
-- `llama3_1-8b`
+- `llama3_1-8b_`
 - `mixtral-8x7b` (base: mixtral-8x7b)
 - `mobilenet`
 - `pointpainting`
@@ -207,4 +207,6 @@ mlcr app,vision,language,mlcommons,mlperf,inference,generic
 - `dlrm_`
 - `gptj_` (alias: gptj)
 - `llama2-70b_`
+- `llama3_1-8b` (base: llama3_1-8b_)
+- `llama3_1-8b-edge` (base: llama3_1-8b_)
 - `power`

script/build-mlperf-inference-server-nvidia/README.md

@@ -1,10 +1,10 @@
 # README for build-mlperf-inference-server-nvidia
 This README is automatically generated. Add custom content in [info.md](info.md). Please follow the [script execution document](https://docs.mlcommons.org/mlcflow/targets/script/execution-flow/) to understand more about the MLC script execution.
 
-`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/user`, you can do
+`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/$USER`, you can do
 ```
-mkdir /mnt/user/MLC
-ln -s /mnt/user/MLC $HOME/MLC
+mkdir /mnt/$USER/MLC
+ln -s /mnt/$USER/MLC $HOME/MLC
 ```
 You can also use the `ENV` variable `MLC_REPOS` to control this location but this will need a set after every system reboot.
 

script/download-file/README.md

@@ -1,10 +1,10 @@
 # README for download-file
 This README is automatically generated. Add custom content in [info.md](info.md). Please follow the [script execution document](https://docs.mlcommons.org/mlcflow/targets/script/execution-flow/) to understand more about the MLC script execution.
 
-`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/user`, you can do
+`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/$USER`, you can do
 ```
-mkdir /mnt/user/MLC
-ln -s /mnt/user/MLC $HOME/MLC
+mkdir /mnt/$USER/MLC
+ln -s /mnt/$USER/MLC $HOME/MLC
 ```
 You can also use the `ENV` variable `MLC_REPOS` to control this location but this will need a set after every system reboot.
 
@@ -72,7 +72,7 @@ mlcr download file
 - `curl`
 - `gdown`
 - `mlcutil` (alias: cmutil) (default)
-- `r2_downloader`
+- `r2-downloader`
 - `rclone`
 - `wget`
 

script/generate-mlperf-inference-submission/README.md

@@ -1,10 +1,10 @@
 # README for generate-mlperf-inference-submission
 This README is automatically generated. Add custom content in [info.md](info.md). Please follow the [script execution document](https://docs.mlcommons.org/mlcflow/targets/script/execution-flow/) to understand more about the MLC script execution.
 
-`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/user`, you can do
+`mlcflow` stores all local data under `$HOME/MLC` by default. So, if there is space constraint on the home directory and you have more space on say `/mnt/$USER`, you can do
 ```
-mkdir /mnt/user/MLC
-ln -s /mnt/user/MLC $HOME/MLC
+mkdir /mnt/$USER/MLC
+ln -s /mnt/$USER/MLC $HOME/MLC
 ```
 You can also use the `ENV` variable `MLC_REPOS` to control this location but this will need a set after every system reboot.