Merge branch 'main' into dev

NOTICES.md

@@ -0,0 +1,67 @@
+# Third-Party Licenses
+
+### python
+- **License:** [Python Software Foundation License](https://docs.python.org/3.8/license.html)
+- **Version:** 3.8.13
+- **Link to Source:** [https://www.python.org/]
+
+### click
+- **License:** [BSD 3-Clause License](https://opensource.org/license/bsd-3-clause/)
+- **Version:** 8.1.7
+- **Link to Source:** [https://github.com/pallets/click]
+
+### pyyaml
+- **License:** [MIT License](https://opensource.org/licenses/MIT)
+- **Version:** 6.0.1
+- **Link to Source:** [https://github.com/yaml/pyyaml]
+
+### asyncer
+- **License:** [MIT License](https://opensource.org/licenses/MIT)
+- **Version:** 0.0.2
+- **Link to Source:** [https://github.com/tiangolo/asyncer]
+
+### jinja2
+- **License:** [BSD 3-Clause License](https://opensource.org/licenses/BSD-3-Clause)
+- **Version:** 3.1.2
+- **Link to Source:** [https://github.com/pallets/jinja]
+
+### python-dotenv
+- **License:** [BSD 3-Clause License](https://opensource.org/licenses/BSD-3-Clause)
+- **Version:** 1.0.0
+- **Link to Source:** [https://github.com/theskumar/python-dotenv]
+
+### cryptography
+- **License:** [BSD 3-Clause License](https://opensource.org/licenses/BSD-3-Clause)
+- **Version:** 41.0.4
+- **Link to Source:** [https://github.com/pyca/cryptography]
+
+### asyncssh
+- **License:** [Eclipse Public License 2.0](https://www.eclipse.org/legal/epl-2.0/)
+- **Version:** 2.14.0
+- **Link to Source:** [https://github.com/ronf/asyncssh]
+
+### bcrypt
+- **License:** [Apache License 2.0](https://opensource.org/licenses/Apache-2.0)
+- **Version:** 3.1.3
+- **Link to Source:** [https://github.com/pyca/bcrypt/]
+
+### libnacl
+- **License:** [Apache License 2.0](https://opensource.org/licenses/Apache-2.0)
+- **Version:** 1.4.2
+- **Link to Source:** [https://github.com/saltstack/libnacl]
+
+### pyopenssl
+- **License:** [Apache License 2.0](https://opensource.org/licenses/Apache-2.0)
+- **Version:** 23.2.0
+- **Link to Source:** [https://github.com/pyca/pyopenssl]
+
+### libsodium
+- **License:** [MIT License](https://opensource.org/licenses/MIT)
+- **Version:** 2.6.1
+- **Link to Source:** [https://github.com/libsodiumproject/sodium]
+
+### invoker
+- **License:** [Apache License 2.0](https://opensource.org/licenses/Apache-2.0)
+- **Version:** latest
+- **Link to Source:** [https://github.com/higgsfield-ai/invoker]
+

higgsfield/internal/cfg.py

@@ -32,10 +32,14 @@ class AppConfig:
  user: str
  key: str
  port: int
+ master_host: Optional[str]
+ no_python: Optional[str]
+ invoker_exec: str = "invoker"
  number_of_processes_per_node: int
 
  def __init__(self, **kwargs):
  self.__dict__.update(kwargs)
+
  @classmethod
  def from_path(cls, path: Path) -> "AppConfig":
  config_path = path / "src" / "config.py"
@@ -66,6 +70,10 @@ def from_path(cls, path: Path) -> "AppConfig":
 
  key = get_key_from_path_or_key(os.getenv("SSH_KEY"))
 
+ master_host = module.__dict__.get("MASTER_HOST", None)
+ no_python = module.__dict__.get("NO_PYTHON", None)
+ invoker_exec = module.__dict__.get("INVOKER_EXEC", "invoker")
+
  return AppConfig(
  name=name,
  github_repo_url=github_repo_url,
@@ -74,6 +82,9 @@ def from_path(cls, path: Path) -> "AppConfig":
  key=key,
  port=port,
  number_of_processes_per_node=number_of_processes_per_node,
+ master_host=master_host,
+ no_python=no_python,
+ invoker_exec=invoker_exec,
  )
 
  def get_git_origin_url(self, path) -> Optional[str]:
@@ -106,7 +117,7 @@ def set_git_origin_url(self, path: Path):
  with open(config_path, "w") as f:
  # Write back the modified lines
  f.writelines(lines)
- 
+
  if lines[-1] != "\n":
  f.write("\n")
 

higgsfield/internal/experiment/builder.py

@@ -105,6 +105,18 @@ def generate(self):
 class ExperimentBuilder(ActionBuilder):
  template_name = "experiment_action.j2"
 
+ def of_master_host(self) -> str:
+ master_host = self.app_config.master_host
+ if master_host is None:
+ return ""
+ return f"--master_host {master_host}"
+
+ def of_no_python(self) -> str:
+ no_python = self.app_config.no_python
+ if no_python is None:
+ return ""
+ return f"--no_python {no_python}"
+
  def generate(self, experiment_name: str, params: List[Param]):
  (self.wf_dir / f"run_{experiment_name}.yml").write_text(
  self.template.render(
@@ -113,12 +125,14 @@ def generate(self, experiment_name: str, params: List[Param]):
  project_name=self.app_config.name,
  params=build_gh_action_inputs(params),
  rest=build_run_params(params),
+ invoker_exec=self.app_config.invoker_exec,
+ of_master_host=self.of_master_host(),
+ of_no_python=self.of_no_python(),
  env_gen=env_keys_as_action(
  self.wf_dir.parent.parent / "env", echo_indent
  ),
  )
  )
-
 
  print("Updated experiment action", experiment_name)
 
@@ -130,7 +144,13 @@ def _source_experiments(base_path: Path):
  the same dependencies (aka environment) as the docker container.
  """
  for file in base_path.glob("**/*.py"):
- module_name = os.path.basename(file).split(".py")[0].split(".py")[0].replace(" ", "_").replace("-", "_")
+ module_name = (
+ os.path.basename(file)
+ .split(".py")[0]
+ .split(".py")[0]
+ .replace(" ", "_")
+ .replace("-", "_")
+ )
  SourceFileLoader(module_name, str(file)).load_module()
 
 

higgsfield/internal/init.py

@@ -76,13 +76,13 @@ def generate_deploy_keys() -> Tuple[bytes, bytes]:
 
  private_bytes = private_key.private_bytes(
  encoding=serialization.Encoding.PEM,
- format=serialization.PrivateFormat.PKCS8,
+ format=serialization.PrivateFormat.OpenSSH,
  encryption_algorithm=serialization.NoEncryption(),
  )
 
  public_bytes = public_key.public_bytes(
- encoding=serialization.Encoding.PEM,
- format=serialization.PublicFormat.SubjectPublicKeyInfo,
+ encoding=serialization.Encoding.OpenSSH,
+ format=serialization.PublicFormat.OpenSSH,
  )
 
  return private_bytes, public_bytes

higgsfield/internal/util.py

@@ -9,8 +9,8 @@
 
 
 def check_name(name: str):
- if len(name) < 1 or len(name) > 20:
- raise ValueError("Name must be between 1 and 20 characters long")
+ if len(name) < 1 or len(name) > 30:
+ raise ValueError("Name must be between 1 and 30 characters long")
 
  if not regex.match(name):
  raise ValueError("Name must match regex ^[a-zA-Z_][a-zA-Z0-9_]*$")

higgsfield/llama/llama.py

@@ -59,14 +59,14 @@ def __init__(
  if not checkpoint_path:
  if cpu_init_rank0: 
  if rank == 0:
- model = LlamaForCausalLM.from_pretrained(model_name)
+ model = LlamaForCausalLM.from_pretrained(model_name, use_cache=False)
  else:
- llama_config = LlamaConfig.from_pretrained(model_name)
+ llama_config = LlamaConfig.from_pretrained(model_name, use_cache=False)
 
  with torch.device('meta'):
  model = LlamaForCausalLM(llama_config)
  else:
- model = LlamaForCausalLM.from_pretrained(model_name)
+ model = LlamaForCausalLM.from_pretrained(model_name, use_cache=False)
  else:
  if not cpu_init_rank0:
  print("Ignoring cpu_init_rank0=False while loading model from checkpoint path")
@@ -298,4 +298,4 @@ def __init__(
  precision,
  cpu_init_rank0,
  cpu_offload,
- )
+ )

higgsfield/static/templates/deploy_action.j2

@@ -37,8 +37,7 @@ jobs:
  host: ${{ env.HOSTS }}
  username: ${{ env.USER }}
  key: ${{ secrets.SSH_KEY }}
- port: ${{ env.PORT }}
- sync: true {% endraw %}
+ port: ${{ env.PORT }} {% endraw %}
  script: | 
  mkdir -p ~/higgsfield/
  cd ~/higgsfield
@@ -48,5 +47,6 @@ jobs:
  git reset --hard origin/main && \
  git pull origin main) || git clone {{ keyed_repo_url }} {{ project_name }} || rm -rf ~/higgsfield/{{ project_name }} && git clone {{ keyed_repo_url }} {{ project_name }}
 
+ cd ~/higgsfield/{{ project_name }}
  echo "SSH_KEY=NOTHING" > env
 {{ env_gen }}

higgsfield/static/templates/experiment_action.j2

@@ -68,7 +68,7 @@ jobs:
  cd ~/higgsfield/{% endraw %}{{ project_name }}
  echo "SSH_KEY=NOTHING" > env
 {{ env_gen }}
- invoker experiment run --project_name {{ project_name }} --experiment_name {{ experiment_name }} {% raw %} --run_name ${{ env.RUN_NAME }} --port ${{ env.RUN_PORT }} --nproc_per_node ${{ env.NPROC_PER_NODE }} --hosts ${{ env.HOSTS }} {% endraw %} {{ rest }}
+ {{ invoker_exec }} experiment run {{ of_master_host }} {{ of_no_python }} --project_name {{ project_name }} --experiment_name {{ experiment_name }} {% raw %} --run_name ${{ env.RUN_NAME }} --port ${{ env.RUN_PORT }} --nproc_per_node ${{ env.NPROC_PER_NODE }} --hosts ${{ env.HOSTS }} {% endraw %} {{ rest }}
 
 
 

pyproject.toml

@@ -14,11 +14,9 @@ pyyaml = "^6.0.1"
 asyncer = "^0.0.2"
 jinja2 = "^3.1.2"
 python-dotenv = "^1.0.0"
-unidecode = "^1.3.7"
 cryptography = "^41.0.4"
-asyncssh = {extras = ["bcrypt", "fido2", "libnacl", "pyopenssl"], version = "^2.14.0"}
+asyncssh = {extras = ["bcrypt", "libnacl", "pyopenssl"], version = "^2.14.0"}
 bcrypt = "^4.0.1"
-fido2 = "^1.1.2"
 libsodium = "^2.6.1"
 pyopenssl = "^23.2.0"