release v0.11.0#2875
Conversation
## Walkthrough
This update removes support for PyTorch 2.5.1 and CUDA 12.4.1 across GitHub Actions workflow matrices, documentation, and default environment variables. The minimum supported PyTorch version is now 2.6.0. Related documentation, tests, and version strings are updated to reflect these changes, with no alterations to core logic or exported entities.
## Changes
| Files/Paths | Change Summary |
|-----------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------|
| .github/workflows/base.yml, main.yml, multi-gpu-e2e.yml, nightlies.yml, | Removed CUDA 12.4.1 + PyTorch 2.5.1 matrix entries; updated matrices to use newer CUDA/PyTorch versions. |
| tests-nightly.yml, tests.yml | |
| README.md, docs/installation.qmd, src/axolotl/integrations/kd/README.md | Updated minimum PyTorch version requirement from 2.5.1 to 2.6.0. |
| docs/docker.qmd | Removed Docker image tags for CUDA 12.4.1 + PyTorch 2.5.1; added/updated tags for CUDA 12.6.3 + PyTorch 2.6.0. |
| cicd/multigpu.py, cicd/single_gpu.py | Changed default PyTorch version to 2.6.0 and CUDA version to 126 in Dockerfile rendering arguments. |
| src/axolotl/__init__.py | Updated version string from "0.11.0.dev" to "0.11.0". |
| src/axolotl/utils/schemas/config.py | Updated torch.compile enablement and warnings to require PyTorch >= 2.6.0. |
| tests/patched/test_validation.py, tests/test_validation_dataset.py | Updated hardcoded torch version in test cases from 2.5.1 to 2.6.0. |
| src/axolotl/monkeypatch/multipack.py | Added "smollm3" to supported multipack model types list. |
## Sequence Diagram(s)
```mermaid
sequenceDiagram
participant User
participant CI Workflow
participant Docker Builder
participant Test Runner
User->>CI Workflow: Pushes code / PR
CI Workflow->>Docker Builder: Build with CUDA 12.6.3/12.8.1, PyTorch 2.6.0/2.7.1
Docker Builder-->>CI Workflow: Returns built images
CI Workflow->>Test Runner: Run tests with PyTorch >=2.6.0
Test Runner-->>CI Workflow: Test results
CI Workflow-->>User: Status (pass/fail)Possibly related PRs
Suggested labels
Suggested reviewers
Poem
|
Codecov ReportAll modified and coverable lines are covered by tests ✅ 📢 Thoughts on this report? Let us know! |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🔭 Outside diff range comments (1)
src/axolotl/utils/schemas/config.py (1)
1058-1066: Fix mismatched version check incheck_torch_compile_autoSame inconsistency – update both occurrences and tighten the warning message:
- ) >= version.parse("2.5.1"): + ) >= version.parse("2.6.0"):Optionally rephrase the log line to “torch.compile is available with torch>=2.6.0”.
♻️ Duplicate comments (5)
tests/test_validation_dataset.py (2)
129-133: Same duplication applies here (and below). A single constant would keep the suite DRY.
185-189: See earlier remark about centralising the literal.tests/patched/test_validation.py (2)
1203-1207: Same comment as above – consider using a shared constant.
1245-1249: Same comment as above – consider using a shared constant..github/workflows/tests.yml (1)
128-129: Same availability concern for the sdist jobThe sdist matrix mirrors the wheel install above – it will hit the same failure mode if 2.7.1 isn’t on PyPI. Ensure both matrices stay in sync once availability is confirmed.
🧹 Nitpick comments (6)
.github/workflows/tests-nightly.yml (1)
29-29: CI matrix cleansed – consider future-proofing.Switching to 2 · 6 · 0 / 2 · 7 · 0 matches the supported range.
Optional: expose the PyTorch list via a reusable YAML anchor/variable so every workflow stays in lock-step when versions change again.tests/test_validation_dataset.py (1)
74-77: Hard-coded torch version duplicatedThe new
"torch_version": "2.6.0"literal is copy-pasted in multiple tests. Consider centralising the minimum-supported version in a helper/fixture to avoid mass edits on the next bump.- env_capabilities={ - "torch_version": "2.6.0", - }, + env_capabilities=MIN_TORCH_ENV_CAPS,tests/patched/test_validation.py (1)
693-697: Prefer constant for torch versionRepeating
"2.6.0"here and in other tests increases drift risk. Export a constant fromtests.conftestor readaxolotl.utils.schemas.config.MIN_TORCH_VERSIONto keep tests aligned with runtime checks.cicd/multigpu.py (1)
27-30: Keep version constants in one place
single_gpu.pyandmultigpu.pynow duplicate"2.6.0","main-base-py3.11-cu126-2.6.0", and"126".
Extract them to a shared module (e.g.cicd/constants.py) to avoid future divergence..github/workflows/main.yml (1)
22-23: Matrix entry now setsaxolotl_extras:to an empty string – consider omitting the keyBecause the subsequent build args/tag logic already no-ops on an empty value, you can delete the field altogether to reduce YAML noise:
- pytorch: 2.6.0 - axolotl_extras: + pytorch: 2.6.0src/axolotl/utils/schemas/config.py (1)
1086-1089: Grammar nit – missing verb in warning
"torch=={torch_version} not be supported"→"torch=={torch_version} is not supported".- f"torch=={torch_version} not be supported. Please upgrade to torch>=2.6.0." + f"torch=={torch_version} is not supported. Please upgrade to torch>=2.6.0."
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (16)
.github/workflows/base.yml(0 hunks).github/workflows/main.yml(1 hunks).github/workflows/multi-gpu-e2e.yml(0 hunks).github/workflows/nightlies.yml(1 hunks).github/workflows/tests-nightly.yml(1 hunks).github/workflows/tests.yml(2 hunks)README.md(1 hunks)cicd/multigpu.py(1 hunks)cicd/single_gpu.py(1 hunks)docs/docker.qmd(1 hunks)docs/installation.qmd(1 hunks)src/axolotl/__init__.py(1 hunks)src/axolotl/integrations/kd/README.md(1 hunks)src/axolotl/utils/schemas/config.py(2 hunks)tests/patched/test_validation.py(3 hunks)tests/test_validation_dataset.py(4 hunks)
💤 Files with no reviewable changes (2)
- .github/workflows/multi-gpu-e2e.yml
- .github/workflows/base.yml
🧰 Additional context used
🧠 Learnings (2)
.github/workflows/tests.yml (1)
Learnt from: NanoCode012
PR: axolotl-ai-cloud/axolotl#2854
File: README.md:73-77
Timestamp: 2025-07-02T02:56:20.788Z
Learning: For Axolotl Docker commands, the `--ipc=host` flag should be included by default to prevent shared memory failures that commonly occur with PyTorch DataLoaders and multiprocessing during machine learning training workflows.
.github/workflows/nightlies.yml (1)
Learnt from: NanoCode012
PR: axolotl-ai-cloud/axolotl#2854
File: README.md:73-77
Timestamp: 2025-07-02T02:56:20.788Z
Learning: For Axolotl Docker commands, the `--ipc=host` flag should be included by default to prevent shared memory failures that commonly occur with PyTorch DataLoaders and multiprocessing during machine learning training workflows.
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (4)
- GitHub Check: docker-e2e-tests (126, 12.6.3, 3.11, 2.7.1, 1)
- GitHub Check: docker-e2e-tests (128, 12.8.1, 3.11, 2.7.1, 1)
- GitHub Check: docker-e2e-tests (126, 12.6.3, 3.11, 2.6.0, 1, llmcompressor)
- GitHub Check: build-base (124, 12.4.1, 3.11, 2.6.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
🔇 Additional comments (9)
README.md (1)
58-58: Update remaining PyTorch 2.5.1 references to 2.6.0We’ve detected stale
2.5.1constraints that need bumping to2.6.0(and>=2.6where appropriate):• src/setuptools_axolotl_dynamic_dependencies.py:48
- torch_version = "2.5.1" + torch_version = "2.6.0"• tests/e2e/utils.py:46, 58, 63, 65
- Decorator marking a test that requires torch >= 2.5.1 + Decorator marking a test that requires torch >= 2.6.0 - return torch_version >= version.parse("2.5.1") + return torch_version >= version.parse("2.6.0") - unittest.skipUnless(is_min_2_5_1(), "test requires torch>=2.5.1") + unittest.skipUnless(is_min_2_6_0(), "test requires torch>=2.6.0")• src/axolotl/utils/schemas/config.py:1031, 1033, 1063
- if version.parse(torch_version) < version.parse("2.5.1"): + if version.parse(torch_version) < version.parse("2.6.0"): - "incompatible with torch version < 2.5.1" + "incompatible with torch version < 2.6.0" - ) >= version.parse("2.5.1"): + ) >= version.parse("2.6.0"):Please update these locations (and any helper function names like
is_min_2_5_1()), and rerun the grep to ensure no old-version strings remain.Likely an incorrect or invalid review comment.
docs/installation.qmd (1)
18-18: Docs updated to ≥2.6.0 – LGTM.The requirement line now aligns with the new baseline.
No further issues spotted in this segment.src/axolotl/integrations/kd/README.md (1)
14-14: Consistent version note – looks good.The comment correctly reflects the new minimum (torch ≥ 2.6.0).
src/axolotl/__init__.py (1)
7-7: Version bump acknowledged – verify packaging metadata is in sync.
__version__is now 0.11.0. Make sure the same number appears in pyproject.toml / setup.cfg to avoid mismatched wheels.docs/docker.qmd (1)
80-83: Docs: new tag looks good, but double-check overall version narrative
main-20250303-py3.11-cu126-2.6.0correctly reflects the 12.6 / 2.6.0 baseline, but earlier in the same section line 58 still says “cuda 12.4”.
Make sure the narrative and all tag examples consistently advertise 12.6 as the new default to avoid user confusion.cicd/single_gpu.py (1)
27-30: Environment defaults updated – ensure Dockerfile template matchesThe hard-coded defaults (
PYTORCH_VERSION,BASE_TAG,CUDA) were bumped.
Double-check thatDockerfile.jinjapicks these vars up (especially the CUDA minor version) and that any CI matrix values are in sync; otherwise local overrides may mask mismatches at build time..github/workflows/tests.yml (2)
55-56: Risk of build failures if torch 2.7.1 is not yet published
pip install torch==${{ matrix.pytorch_version }}will fail outright if 2.7.1 has not landed on PyPI at the time the workflow runs.
Double-check that 2.7.1 is actually available; otherwise pin to the latest released 2.7.x nightly URL or stick with 2.6.0 for now.
299-307: Left-over CUDA 12.4.1 cleanup image contradicts the “drop 12.4.1” objective
docker-e2e-cleanupstill builds a 12.4.1 image (now with torch 2.6.0). If the goal is to eliminate 12.4.1 entirely, this entry should be removed; if it’s intentional (cleaning up legacy artefacts) please add a comment explaining its retention..github/workflows/nightlies.yml (1)
66-72: 👍 Nightly matrix bumped to torch 2.6.0Switching the 12.4.1 entry to torch 2.6.0 aligns with the new minimum version and should keep nightly builds green. No further issues spotted here.
| "description": "Whether to use torch.compile and which backend to use. setting to `auto` will enable torch compile when torch>=2.6.0" | ||
| }, |
There was a problem hiding this comment.
Docstring updated but validation logic still checks for ≥ 2.5.1
The check_torch_compile_auto validator below still enables torch_compile when torch ≥ 2.5.1, contradicting the new requirement (≥ 2.6.0) stated here. Align the threshold:
- ) >= version.parse("2.5.1"):
+ ) >= version.parse("2.6.0"):📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| "description": "Whether to use torch.compile and which backend to use. setting to `auto` will enable torch compile when torch>=2.6.0" | |
| }, | |
| @validator('torch_compile', pre=True) | |
| def check_torch_compile_auto(cls, v): | |
| if v == 'auto': | |
| - if ( | |
| - version.parse(torch.__version__) | |
| - ) >= version.parse("2.5.1"): | |
| + if ( | |
| + version.parse(torch.__version__) | |
| + ) >= version.parse("2.6.0"): | |
| return True | |
| return False | |
| return v |
🤖 Prompt for AI Agents
In src/axolotl/utils/schemas/config.py around lines 630 to 631, the docstring
states that torch.compile should be enabled for torch versions ≥ 2.6.0, but the
validator check_torch_compile_auto still uses ≥ 2.5.1 as the threshold. Update
the validation logic in check_torch_compile_auto to use torch version 2.6.0 as
the minimum required version to enable torch_compile, ensuring consistency with
the docstring.
There was a problem hiding this comment.
Do we need the version check anymore?
There was a problem hiding this comment.
🧩 Analysis chain
Let me investigate the codebase to better understand the usage patterns and determine if the version check is still necessary.
🏁 Script executed:
#!/bin/bash
# Search for torch_compile usage patterns in the codebase
echo "=== Searching for torch_compile usage patterns ==="
rg "torch_compile" -A 3 -B 3
echo -e "\n=== Checking for other torch version checks ==="
rg "version\.parse.*torch" -A 2 -B 2
echo -e "\n=== Looking for torch.compile actual usage ==="
rg "torch\.compile" -A 2 -B 2
Length of output: 24617
@salmanmohammadi Based on the codebase analysis, you raise a valid question. Let me examine the current situation:
Current state:
- Minimum PyTorch version is now 2.6.0 (as updated in this PR)
torch.compileis available in PyTorch 2.6.0- There's already a general minimum version check (
check_min_torch_version) that warns users about torch < 2.6.0
The version check in check_torch_compile_auto could be:
- Simplified - Since the minimum version is 2.6.0, we could simplify the logic:
@model_validator(mode="before")
@classmethod
def check_torch_compile_auto(cls, data):
if data.get("torch_compile") == "auto":
LOG.info("torch.compile is available, setting torch_compile to True")
data["torch_compile"] = True
return data- Removed entirely - Since torch.compile should always be available in supported versions,
torch_compile: "auto"could just be treated astorch_compile: Truewithout any conditional logic.
However, there are reasons to keep it:
- Safety net: Protects against edge cases where someone might run with an unsupported version
- Explicit feedback: Provides clear logging about why torch.compile is/isn't enabled
- Consistency: Other features like
check_flex_torch_versionandcheck_qat_configstill have version checks even at the minimum version threshold
My recommendation: Update the version check to use 2.6.0 (matching the minimum requirement) rather than removing it entirely. This maintains the safety net while being consistent with the new minimum version:
) >= version.parse("2.6.0"):What's your preference? Would you like to simplify/remove the check entirely, or just update it to 2.6.0?
🧬 Code Graph Analysis Results
Relevant snippets from src/axolotl/utils/schemas/config.py (lines 300-420 approx)
@model_validator(mode="before")
@classmethod
def check_adopt_torch_version(cls, data):
if (data.get("optimizer") is not None) and ("adopt" in data.get("optimizer")):
env_capabilities = data.get("env_capabilities", {})
torch_version = env_capabilities.get("torch_version")
if torch_version is None:
import torch
torch_version = str(torch.__version__).split("+", maxsplit=1)[0]
if version.parse(torch_version) < version.parse("2.5.1"):
raise ValueError(
"ADOPT optimizer is incompatible with torch version < 2.5.1"
)
return data
@model_validator(mode="before")
@classmethod
def check_flex_torch_version(cls, data):
if (data.get("flex_attention") is not None) and (data.get("flex_attention")):
env_capabilities = data.get("env_capabilities", {})
torch_version = env_capabilities.get("torch_version")
if torch_version is None:
import torch
torch_version = str(torch.__version__).split("+", maxsplit=1)[0]
if version.parse(torch_version) < version.parse("2.6.0"):
raise ValueError(
"Flex attention is not supported on torch version < 2.6.0"
)
return data
@model_validator(mode="before")
@classmethod
def check_torch_compile_auto(cls, data):
if data.get("torch_compile") == "auto":
env_capabilities = data.get("env_capabilities", {})
if env_capabilities.get("torch_version"):
if version.parse(
env_capabilities.get("torch_version")
) >= version.parse("2.5.1"):
LOG.info(
"torch.compile is available, setting torch_compile to True"
)
data["torch_compile"] = True
else:
data["torch_compile"] = False
else:
data["torch_compile"] = False
return data
@model_validator(mode="after")
def check_min_torch_version(self):
if self.env_capabilities and self.env_capabilities.torch_version:
torch_version = self.env_capabilities.torch_version
if version.parse(torch_version) < version.parse("2.6.0"):
LOG.warning(
f"torch=={torch_version} not be supported. Please upgrade to torch>=2.6.0."
)
return selfThese validators perform version checks on the PyTorch version in the environment capabilities and raise errors or warnings if the version is incompatible with certain features or configurations (e.g., ADOPT optimizer requires torch >= 2.5.1, flex_attention requires torch >= 2.6.0, torch.compile auto-enables only if torch >= 2.5.1, and a warning is logged if torch version is below 2.6.0).
Relevant snippet from src/axolotl/utils/schemas/internal/__init__.py (lines 8-21)
class GPUCapabilities(BaseModel):
"""model to manage the gpu capabilities statically"""
bf16: bool = Field(default=False)
fp8: bool = Field(default=False)
n_gpu: int = Field(default=1)
n_node: int = Field(default=1)
compute_capability: Optional[str] = Field(default=None)
class EnvCapabilities(BaseModel):
"""model to manage the environment capabilities statically"""
torch_version: Optional[str] = Field(default=None)These classes define the environment and GPU capabilities models that are used in the version checks in the config validators.
Relevant snippet from src/axolotl/utils/schemas/deprecated.py (lines 12-56)
class DeprecatedParameters(BaseModel):
"""configurations that are deprecated"""
max_packed_sequence_len: int | None = None
rope_scaling: Any | None = None
noisy_embedding_alpha: float | None = None
dpo_beta: float | None = None
evaluation_strategy: str | None = None
@field_validator("max_packed_sequence_len")
@classmethod
def validate_max_packed_sequence_len(cls, max_packed_sequence_len):
if max_packed_sequence_len:
raise DeprecationWarning("`max_packed_sequence_len` is no longer supported")
return max_packed_sequence_len
@field_validator("rope_scaling")
@classmethod
def validate_rope_scaling(cls, rope_scaling):
if rope_scaling:
raise DeprecationWarning(
"`rope_scaling` is no longer supported, it should now be be a key under `model_config`"
)
return rope_scaling
@field_validator("noisy_embedding_alpha")
@classmethod
def validate_noisy_embedding_alpha(cls, noisy_embedding_alpha):
if noisy_embedding_alpha:
LOG.warning("noisy_embedding_alpha is deprecated, use neftune_noise_alpha")
return noisy_embedding_alpha
@field_validator("dpo_beta")
@classmethod
def validate_dpo_beta(cls, dpo_beta):
if dpo_beta is not None:
LOG.warning("dpo_beta is deprecated, use rl_beta instead")
return dpo_beta
@field_validator("evaluation_strategy")
@classmethod
def validate_evaluation_strategy(cls, evaluation_strategy):
if evaluation_strategy is not None:
LOG.warning("evaluation_strategy is deprecated, use eval_strategy instead")
return evaluation_strategyThis snippet shows deprecated parameters and their validators that raise warnings or errors if used. This is related to versioning and backward compatibility.
Summary of relevance:
- The version checks in
AxolotlConfigWCapabilitiesclass validators (especiallycheck_adopt_torch_version,check_flex_torch_version,check_torch_compile_auto, andcheck_min_torch_version) are the main places where version checking logic is implemented. - The environment capabilities model (
EnvCapabilities) provides the torch version info used in these checks. - Deprecated parameters and their validators in
DeprecatedParametersclass also relate to versioning and backward compatibility. - No other snippets explicitly show version checks or related logic.
These snippets provide the context needed to evaluate whether the version check is still necessary or can be removed.
| - NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU | ||
| - Python 3.11 | ||
| - PyTorch ≥2.5.1 | ||
| - PyTorch ≥2.6.0 |
There was a problem hiding this comment.
Actionable comments posted: 1
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (17)
.github/workflows/base.yml(0 hunks).github/workflows/main.yml(1 hunks).github/workflows/multi-gpu-e2e.yml(0 hunks).github/workflows/nightlies.yml(1 hunks).github/workflows/tests-nightly.yml(1 hunks).github/workflows/tests.yml(3 hunks)README.md(1 hunks)cicd/multigpu.py(1 hunks)cicd/single_gpu.py(1 hunks)docs/docker.qmd(1 hunks)docs/installation.qmd(1 hunks)src/axolotl/__init__.py(1 hunks)src/axolotl/integrations/kd/README.md(1 hunks)src/axolotl/monkeypatch/multipack.py(1 hunks)src/axolotl/utils/schemas/config.py(2 hunks)tests/patched/test_validation.py(3 hunks)tests/test_validation_dataset.py(4 hunks)
💤 Files with no reviewable changes (2)
- .github/workflows/multi-gpu-e2e.yml
- .github/workflows/base.yml
✅ Files skipped from review due to trivial changes (1)
- docs/docker.qmd
🚧 Files skipped from review as they are similar to previous changes (13)
- docs/installation.qmd
- src/axolotl/integrations/kd/README.md
- src/axolotl/monkeypatch/multipack.py
- cicd/multigpu.py
- src/axolotl/init.py
- cicd/single_gpu.py
- tests/test_validation_dataset.py
- .github/workflows/main.yml
- .github/workflows/tests-nightly.yml
- src/axolotl/utils/schemas/config.py
- .github/workflows/nightlies.yml
- tests/patched/test_validation.py
- .github/workflows/tests.yml
🧰 Additional context used
🪛 LanguageTool
README.md
[grammar] ~58-~58: There might be a problem here.
Context: ...ion) or AMD GPU - Python 3.11 - PyTorch ≥2.6.0 ### Installation #### Using pip ```bash pip3 install -U pack...
(QB_NEW_EN_MERGED_MATCH)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (17)
- GitHub Check: PyTest from Source Dist (3.11, 2.7.1)
- GitHub Check: PyTest (3.11, 2.7.0)
- GitHub Check: PyTest from Source Dist (3.11, 2.6.0)
- GitHub Check: PyTest (3.11, 2.7.1)
- GitHub Check: PyTest from Source Dist (3.11, 2.7.0)
- GitHub Check: PyTest (3.11, 2.6.0)
- GitHub Check: build-base (128, 12.8.1, 3.11, nightly, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base-nightly)
- GitHub Check: pre-commit
- GitHub Check: pre-commit
- GitHub Check: build-base (126, 12.6.3, 3.11, 2.7.1, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
- GitHub Check: build-base (126, 12.6.3, 3.11, 2.6.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
- GitHub Check: build-base (126, 12.6.3, 3.11, 2.7.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
- GitHub Check: build-base (128, 12.6.3, 3.11, 2.7.1, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
- GitHub Check: build-base (124, 12.4.1, 3.11, 2.6.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-base)
- GitHub Check: build-base-uv (128, 12.8.1, 3.11, 2.7.1, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-uv-base)
- GitHub Check: preview
- GitHub Check: build-base-uv (126, 12.6.3, 3.11, 2.6.0, 7.0 7.5 8.0 8.6 8.7 8.9 9.0+PTX, Dockerfile-uv-base)
| - NVIDIA GPU (Ampere or newer for `bf16` and Flash Attention) or AMD GPU | ||
| - Python 3.11 | ||
| - PyTorch ≥2.5.1 | ||
| - PyTorch ≥2.6.0 |
There was a problem hiding this comment.
💡 Verification agent
🧩 Analysis chain
Ensure all residual references to PyTorch 2.5.1 are removed across the repo
The bump to “PyTorch ≥ 2.6.0” looks correct here, but stray mentions of 2.5.1 in docs, scripts, or CI can confuse users.
Run the following to confirm there are no leftover references:
🏁 Script executed:
#!/bin/bash
# Search the entire repository for any remaining 2.5.1 mentions
rg -n "2\.5\.1"Length of output: 773
Update all residual PyTorch 2.5.1 references to ≥ 2.6.0
The bump in README.md is correct, but we still have a few stray “2.5.1” mentions that need updating:
- src/setuptools_axolotl_dynamic_dependencies.py (line 48)
– Changetotorch_version = "2.5.1"
torch_version = "2.6.0"
- tests/e2e/utils.py
– Doc comments on lines 46 & 58 referencing “torch ≥ 2.5.1”
– Version check on line 63:– Skip message on line 65:return torch_version >= version.parse("2.5.1")
Update all to use “2.6.0” (or bump threshold to 2.6.0).unittest.skipUnless(is_min_2_5_1(), "test requires torch>=2.5.1")
- src/axolotl/utils/schemas/config.py (lines 1031, 1033, 1063)
– Comparisons and error text usingversion.parse("2.5.1")→version.parse("2.6.0")
After making these changes, please rerun
rg -n "2\.5\.1"to ensure no leftovers.
🧰 Tools
🪛 LanguageTool
[grammar] ~58-~58: There might be a problem here.
Context: ...ion) or AMD GPU - Python 3.11 - PyTorch ≥2.6.0 ### Installation #### Using pip ```bash pip3 install -U pack...
(QB_NEW_EN_MERGED_MATCH)
🤖 Prompt for AI Agents
In README.md lines 56 to 58 and additionally in
src/setuptools_axolotl_dynamic_dependencies.py line 48, tests/e2e/utils.py lines
46, 58, 63, and 65, and src/axolotl/utils/schemas/config.py lines 1031, 1033,
and 1063, update all references of PyTorch version "2.5.1" to "2.6.0". This
includes string assignments, version comparisons, doc comments, and skip
messages. After these changes, run a search for "2.5.1" to confirm no references
remain.
3 participants
Summary by CodeRabbit
Documentation
Chores
Tests
Refactor
Style
New Features