truss 0.11.7__tar.gz → 0.11.8__tar.gz

{truss-0.11.7 → truss-0.11.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truss
-Version: 0.11.7
+Version: 0.11.8
 Summary: A seamless bridge from model development to model delivery
 Project-URL: Repository, https://github.com/basetenlabs/truss
 Project-URL: Homepage, https://truss.baseten.co

{truss-0.11.7 → truss-0.11.8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "truss"
-version = "0.11.7"
+version = "0.11.8"
 description = "A seamless bridge from model development to model delivery"
 authors = [
     { name = "Pankaj Gupta", email = "no-reply@baseten.co" },

{truss-0.11.7 → truss-0.11.8}/truss/base/trt_llm_config.py

@@ -329,9 +329,16 @@ pip install truss==0.10.8
             raise ValueError("Using fp8 context fmha requires paged context fmha")
         if (
             self.plugin_configuration.use_fp8_context_fmha
-            and not self.quantization_type == TrussTRTLLMQuantizationType.FP8_KV
+            and self.quantization_type
+            not in (
+                TrussTRTLLMQuantizationType.FP8_KV,
+                TrussTRTLLMQuantizationType.FP4_KV,
+            )
         ):
-            raise ValueError("Using fp8 context fmha requires fp8 kv cache dtype")
+            raise ValueError(
+                "Using fp8 context fmha requires fp8 kv, or fp4 with kv cache dtype"
+            )
+
         return self
 
     def _validate_speculator_config(self):

{truss-0.11.7 → truss-0.11.8}/truss/cli/train/core.py

@@ -156,7 +156,7 @@ def display_training_projects(projects: list[dict], remote_url: str) -> None:
         latest_job = project.get("latest_job") or {}
         if latest_job_id := latest_job.get("id", ""):
             latest_job_link = cli_common.format_link(
-                status_page_url(remote_url, latest_job_id), "link"
+                status_page_url(remote_url, project["id"], latest_job_id), "link"
             )
         else:
             latest_job_link = ""
@@ -305,8 +305,11 @@ def display_training_job(
     table.add_row("Created", cli_common.format_localized_time(job["created_at"]))
     table.add_row("Last Modified", cli_common.format_localized_time(job["updated_at"]))
     table.add_row(
-        "Status Page",
-        cli_common.format_link(status_page_url(remote_url, job["id"]), "link"),
+        "Job Page",
+        cli_common.format_link(
+            status_page_url(remote_url, job["training_project"]["id"], job["id"]),
+            "link",
+        ),
     )
 
     # Add error message if present
@@ -417,8 +420,8 @@ def download_checkpoint_artifacts(
     return urls_file
 
 
-def status_page_url(remote_url: str, training_job_id: str) -> str:
-    return f"{remote_url}/training/jobs/{training_job_id}"
+def status_page_url(remote_url: str, project_id: str, training_job_id: str) -> str:
+    return f"{remote_url}/training/{project_id}/logs/{training_job_id}"
 
 
 def _get_all_train_init_example_options(

{truss-0.11.7 → truss-0.11.8}/truss/cli/train_commands.py

@@ -41,13 +41,14 @@ truss_cli.add_command(train)
 
 def _print_training_job_success_message(
     job_id: str,
+    project_id: str,
     project_name: str,
-    job_object: TrainingJob,
+    job_object: Optional[TrainingJob],
     remote_provider: BasetenRemote,
 ) -> None:
     """Print success message and helpful commands for a training job."""
     console.print("✨ Training job successfully created!", style="green")
-    should_print_cache_summary = (
+    should_print_cache_summary = job_object and (
         job_object.runtime.enable_cache
         or job_object.runtime.cache_config
         and job_object.runtime.cache_config.enabled
@@ -64,7 +65,7 @@ def _print_training_job_success_message(
         f"🔍 View metrics for your job via "
         f"[cyan]'truss train metrics --job-id {job_id}'[/cyan]\n"
         f"{cache_summary_snippet}"
-        f"🌐 Status page: {common.format_link(core.status_page_url(remote_provider.remote_url, job_id))}"
+        f"🌐 View job in the UI: {common.format_link(core.status_page_url(remote_provider.remote_url, project_id, job_id))}"
     )
 
 
@@ -80,8 +81,13 @@ def _handle_post_create_logic(
             style="green",
         )
     else:
+        # recreate currently doesn't pass back a job object.
         _print_training_job_success_message(
-            job_id, project_name, job_resp["job_object"], remote_provider
+            job_id,
+            project_id,
+            project_name,
+            job_resp.get("job_object"),
+            remote_provider,
         )
 
     if tail:
@@ -156,11 +162,16 @@ def recreate_training_job(job_id: Optional[str], remote: Optional[str], tail: bo
 @train.command(name="logs")
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option("--tail", is_flag=True, help="Tail for ongoing logs.")
 @common.common_options()
 def get_job_logs(
-    remote: Optional[str], project_id: Optional[str], job_id: Optional[str], tail: bool
+    remote: Optional[str],
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    tail: bool,
 ):
     """Fetch logs for a training job"""
 
@@ -170,6 +181,10 @@ def get_job_logs(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
+
     project_id, job_id = train_common.get_most_recent_job(
         remote_provider, project_id, job_id
     )
@@ -188,12 +203,17 @@ def get_job_logs(
 
 @train.command(name="stop")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option("--all", is_flag=True, help="Stop all running jobs.")
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common.common_options()
 def stop_job(
-    project_id: Optional[str], job_id: Optional[str], all: bool, remote: Optional[str]
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    all: bool,
+    remote: Optional[str],
 ):
     """Stop a training job"""
 
@@ -203,6 +223,9 @@ def stop_job(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
     if all:
         train_cli.stop_all_jobs(remote_provider, project_id)
     else:
@@ -217,13 +240,17 @@ def stop_job(
 @click.option(
     "--project-id", type=str, required=False, help="View training jobs for a project."
 )
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option(
     "--job-id", type=str, required=False, help="View a specific training job."
 )
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common.common_options()
 def view_training(
-    project_id: Optional[str], job_id: Optional[str], remote: Optional[str]
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    remote: Optional[str],
 ):
     """List all training jobs for a project"""
 
@@ -233,16 +260,24 @@ def view_training(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
+
     train_cli.view_training_details(remote_provider, project_id, job_id)
 
 
 @train.command(name="metrics")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common.common_options()
 def get_job_metrics(
-    project_id: Optional[str], job_id: Optional[str], remote: Optional[str]
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    remote: Optional[str],
 ):
     """Get metrics for a training job"""
 
@@ -252,11 +287,15 @@ def get_job_metrics(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
     train_cli.view_training_job_metrics(remote_provider, project_id, job_id)
 
 
 @train.command(name="deploy_checkpoints")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option(
     "--config",
@@ -271,6 +310,7 @@ def get_job_metrics(
 @common.common_options()
 def deploy_checkpoints(
     project_id: Optional[str],
+    project: Optional[str],
     job_id: Optional[str],
     config: Optional[str],
     remote: Optional[str],
@@ -286,6 +326,9 @@ def deploy_checkpoints(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
     prepare_checkpoint_result = train_cli.prepare_checkpoint_deploy(
         remote_provider,
         train_cli.PrepareCheckpointArgs(
@@ -492,3 +535,15 @@ def view_cache_summary(project: str, remote: Optional[str], sort: str, order: st
     )
 
     train_cli.view_cache_summary_by_project(remote_provider, project, sort, order)
+
+
+def _maybe_resolve_project_id_from_id_or_name(
+    remote_provider: BasetenRemote, project_id: Optional[str], project: Optional[str]
+) -> Optional[str]:
+    """resolve the project_id or project. `project` can be name or id"""
+    if project and project_id:
+        console.print("Both `project-id` and `project` provided. Using `project`.")
+    project_str = project or project_id
+    if not project_str:
+        return None
+    return train_cli.fetch_project_by_name_or_id(remote_provider, project_str)["id"]

{truss-0.11.7 → truss-0.11.8}/truss/templates/base.Dockerfile.jinja

@@ -33,7 +33,7 @@ RUN useradd -u {{ app_user_uid }} -ms /bin/bash {{ app_username }}
 ENV DEBIAN_FRONTEND=noninteractive
 
 
-{%- set UV_VERSION = "0.7.19" %}
+{%- set UV_VERSION = "0.8.22" %}
 {#
 NB(nikhil): We use a semi-complex uv installation command across the board:
 - A generous UV_HTTP_TIMEOUT (5m) for packages that take a long time to install.
@@ -59,10 +59,12 @@ RUN {{ python_exec_path }} -c "import sys; \
 {% endblock %}
 
 {% block install_uv %}
-{# Install `uv` and `curl` if not already present in the image. #}
+{# Install `uv` and `curl` if not already present in the image. We validate the expected location for `uv` at the very end
+   due to limitations with `pipefail` in Docker context. #}
 RUN if ! command -v uv >/dev/null 2>&1; then \
     command -v curl >/dev/null 2>&1 || (apt update && apt install -y curl) && \
-    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/{{ UV_VERSION }}/install.sh | sh; \
+    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/{{ UV_VERSION }}/install.sh | sh && \
+    test -x ${HOME}/.local/bin/uv; \
 fi
 {# Add the user's local bin to the path, used by uv. #}
 ENV PATH=${PATH}:${HOME}/.local/bin

{truss-0.11.7 → truss-0.11.8}/truss/templates/docker_server/proxy.conf.jinja

@@ -45,6 +45,19 @@ server {
         proxy_pass http://127.0.0.1:{{server_port}};
     }
 
+    location ~ ^/v1/websocket$ {
+        proxy_redirect off;
+        proxy_read_timeout 18030s;
+        proxy_http_version 1.1;
+
+        proxy_set_header Upgrade $upgrade_header;
+        proxy_set_header Connection $connection_header;
+
+        rewrite ^/v1/websocket$ {{server_endpoint}} break;
+
+        proxy_pass http://127.0.0.1:{{server_port}};
+    }
+
     # Forward all other paths
     location / {
         proxy_redirect off;

{truss-0.11.7 → truss-0.11.8}/truss/templates/server/requirements.txt

@@ -18,7 +18,7 @@ psutil>=5.9.4
 python-json-logger>=2.0.2
 pyyaml>=6.0.0
 requests>=2.31.0
-truss-transfer==0.0.31
+truss-transfer==0.0.32
 # NB(nikhil): Uvicorn 0.36.0 has breaking changes for the event loop, so we pin to a lower version.
 uvicorn>=0.24.0,<0.36.0
 uvloop>=0.19.0

{truss-0.11.7 → truss-0.11.8}/truss/tests/test_data/server.Dockerfile

@@ -17,7 +17,8 @@ RUN /usr/local/bin/python3 -c "import sys; \
     || { echo "ERROR: Supplied base image does not have 3.8 <= python <= 3.13"; exit 1; }
 RUN if ! command -v uv >/dev/null 2>&1; then \
     command -v curl >/dev/null 2>&1 || (apt update && apt install -y curl) && \
-    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/0.7.19/install.sh | sh; \
+    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/0.8.22/install.sh | sh && \
+    test -x ${HOME}/.local/bin/uv; \
 fi
 ENV PATH=${PATH}:${HOME}/.local/bin
 ENV PYTHONUNBUFFERED="True"

{truss-0.11.7 → truss-0.11.8}/truss-transfer/README.md

@@ -264,15 +264,47 @@ The following environment variables can be used to configure truss-transfer beha
 
 ### Core Configuration
 
-- **`TRUSS_TRANSFER_DOWNLOAD_DIR`** (default: `/tmp/bptr-resolved`)
+- **`TRUSS_TRANSFER_DOWNLOAD_DIR`** (default: `/tmp/truss_transfer`)
   - Directory where resolved files will be downloaded
   - Used when no explicit download directory is provided
   - Can be overridden by passing a directory to the CLI or Python function
 
-- **`RUST_LOG`** (default: `info`)
+- **`TRUSS_TRANSFER_LOG`** or **`RUST_LOG`** (default: `info`)
   - Controls logging level: `error`, `warn`, `info`, `debug`, `trace`
+  - `TRUSS_TRANSFER_LOG` takes precedence over `RUST_LOG`
   - Example: `RUST_LOG=debug` for detailed logging
 
+- **`TRUSS_TRANSFER_CACHE_DIR`** (default: `/cache/org/artifacts/truss_transfer_managed_v1`)
+  - Cache directory for b10fs operations
+  - Used when Baseten FS is enabled
+
+### Download Configuration
+
+- **`TRUSS_TRANSFER_NUM_WORKERS`** (default: `6`)
+  - Number of concurrent download workers
+  - Controls parallelism for file downloads
+
+- **`TRUSS_TRANSFER_USE_RANGE_DOWNLOAD`** (default: `false`)
+  - Enable/disable range-based downloading for large files
+  - Set to `1`, `true`, `yes`, or `y` to enable
+
+- **`TRUSS_TRANSFER_RANGE_DOWNLOAD_WORKERS`** (default: `192`)
+  - Total number of range download workers across all files
+  - Used when range downloading is enabled
+
+- **`TRUSS_TRANSFER_RANGE_DOWNLOAD_WORKERS_PER_FILE`** (default: `84`)
+  - Number of concurrent range workers per individual file
+  - Used when range downloading is enabled
+
+- **`TRUSS_TRANSFER_DOWNLOAD_MONITOR_SECS`** (default: `30`)
+  - Interval in seconds for monitoring download progress
+  - Controls how often progress is reported
+
+- **`TRUSS_TRANSFER_PAGE_AFTER_DOWNLOAD`** (default: `false`)
+  - Enable/disable memory paging after downloads complete
+  - Set to `1`, `true`, `yes`, or `y` to enable
+  - Helps with memory management for large downloads
+
 ### Authentication
 
 - **`HF_TOKEN`** (optional)
@@ -289,29 +321,46 @@ The following environment variables can be used to configure truss-transfer beha
 
 - **`BASETEN_FS_ENABLED`** (default: `false`)
   - Enable/disable Baseten FS caching: `1`/`true` to enable, `0`/`false` to disable
-  - When enabled, files are cached in `/cache/org/artifacts/truss_transfer_managed_v1`
+  - When enabled, files are cached in the directory specified by `TRUSS_TRANSFER_CACHE_DIR`
 
 - **`TRUSS_TRANSFER_B10FS_CLEANUP_HOURS`** (default: `96`)
   - Hours after last access before deleting cached files from other tenants
   - Helps manage disk space by removing old cached files
   - Example: `TRUSS_TRANSFER_B10FS_CLEANUP_HOURS=48` for 2 days
 
-- **`TRUSS_TRANSFER_B10FS_DOWNLOAD_SPEED_MBPS`** (default: `350`)
+- **`TRUSS_TRANSFER_B10FS_DOWNLOAD_SPEED_MBPS`** (default: dynamic)
   - Expected download speed in MB/s for b10fs performance benchmarking
   - Used to determine if b10fs is faster than direct download
+  - Default: 400 MB/s for >16 cores, 90 MB/s for ≤16 cores (with randomization)
   - Lower values make b10fs more likely to be used
 
+- **`TRUSS_TRANSFER_B10FS_MAX_STALE_CACHE_SIZE_GB`** (default: unlimited)
+  - Maximum size in GB for stale cache files before cleanup is triggered
+  - When set, actively purges old cache files to maintain this limit
+  - Example: `TRUSS_TRANSFER_B10FS_MAX_STALE_CACHE_SIZE_GB=500`
+
 ### Example Configuration
 
 ```bash
 # Basic setup
 export TRUSS_TRANSFER_DOWNLOAD_DIR="/tmp/my-models"
-export RUST_LOG=info
+export TRUSS_TRANSFER_LOG=info
+export TRUSS_TRANSFER_NUM_WORKERS=8
 
-# With b10fs enabled and authentication
+# Advanced download configuration
+export TRUSS_TRANSFER_USE_RANGE_DOWNLOAD=1
+export TRUSS_TRANSFER_RANGE_DOWNLOAD_WORKERS=256
+export TRUSS_TRANSFER_RANGE_DOWNLOAD_WORKERS_PER_FILE=64
+export TRUSS_TRANSFER_PAGE_AFTER_DOWNLOAD=1
+
+# With b10fs enabled and tuned
 export BASETEN_FS_ENABLED=1
+export TRUSS_TRANSFER_CACHE_DIR="/fast-ssd/cache"
 export TRUSS_TRANSFER_B10FS_CLEANUP_HOURS=48
-export TRUSS_TRANSFER_B10FS_DOWNLOAD_SPEED_MBPS=100
+export TRUSS_TRANSFER_B10FS_DOWNLOAD_SPEED_MBPS=200
+export TRUSS_TRANSFER_B10FS_MAX_STALE_CACHE_SIZE_GB=1000
+
+# Authentication
 export HF_TOKEN="your-huggingface-token"
 ```