Bump filelock from 3.20.2 to 3.20.3#1
Open
dependabot[bot] wants to merge 1 commit into
Open
Conversation
Bumps [filelock](https://github.com/tox-dev/py-filelock) from 3.20.2 to 3.20.3. - [Release notes](https://github.com/tox-dev/py-filelock/releases) - [Changelog](https://github.com/tox-dev/filelock/blob/main/docs/changelog.rst) - [Commits](tox-dev/filelock@3.20.2...3.20.3) --- updated-dependencies: - dependency-name: filelock dependency-version: 3.20.3 dependency-type: direct:production ... Signed-off-by: dependabot[bot] <support@github.com>
dependabot
Bot
added
dependencies
CityChan
added a commit
to CityChan/MemexRL
that referenced
this pull request
May 7, 2026
…arm-up generator
Three additions to the graph_db memory mode aimed at the three biggest cold-start
risks identified during review: edge type vocabulary drift, sparse training signal,
and unprepared base-model output distribution. All additions are strictly additive
and default-off where they would change baseline behaviour.
1. Closed edge schema (graph_context_database.py / mixin.py / prompts.py /
tool_agent.py / __init__.py)
- GraphContextDatabase accepts an optional edge_schema. Edge types are
lowercased on insert; types outside the schema are soft-dropped and
counted in _dropped_edge_count for telemetry.
- QueryGraph edge_types filter is also lowercase-canonicalized so callers
don't have to match case.
- add_edge() now respects the schema, returns success bool, and registers
endpoints in the entity index so cold-start linkers' nodes show up in
list_entities().
- get_memory_tools_prompt_graph(edge_schema=...) renders the closed
vocabulary into the prompt so the policy doesn't have to invent edge
labels. ALFWORLD_EDGE_SCHEMA and HOTPOTQA_EDGE_SCHEMA presets exported.
- ToolAgentWithMemory and init_memory thread edge_schema through to the
factory.
2. Retrieval utility reward (memory_efficiency_shaper.py)
- New MemoryEfficiencyShaper term: when a ReadExperience or QueryGraph call
surfaces a key (db_index or focus entity), reward shaping fires iff that
key appears within util_window_steps (default 5) in either a non-memory
tool call's arguments or the model's response text.
- QueryGraph added to MEMORY_TOOLS so memory-tool chaining can't farm bonus.
RETRIEVAL_TOOLS subset added.
- Off by default (enable_utility_reward=False) so existing runs are
bit-identical.
3. SFT warm-up data generator (scripts/build_graph_sft_data.py)
- Converts lossless_db trajectory JSONL into graph_db format by extracting
entity / entities / relations on each db_block in CompressExperience
calls. Two extractor backends: a deterministic MockExtractor for tests
and an OpenAIExtractor for real warm-up data prep. Bad LLM JSON falls
back to mock per-block so the pipeline never aborts.
- Aimed at the cold-start step Accenture#1 in docs/graph_memory.md: give the base
model an SFT prior over graph schema before RL training.
Tests: 22 new (10 retrieval-utility + 12 SFT generator) on top of 37 existing
graph tests. Total 59/59 pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
CityChan
added a commit
to CityChan/MemexRL
that referenced
this pull request
May 8, 2026
Two scripts to run on a single GH200 idev right after install_slime_vista.sh
finishes, before committing to multi-node 30B training.
scripts/smoke_4b_inference.py
Pure inference smoke. Validates: CUDA driver, transformers wheel,
Slime stack imports (flash_attn, transformer_engine, apex, megatron,
sglang, ring_flash_attn), 4B model load, forward pass, generate 50
tokens. Catches "is the install actually functional?" without
pulling in any MemexRL agent code or ALFWorld dependencies.
scripts/smoke_4b_alfworld.py
End-to-end agent + env smoke. Drives one ALFWorld episode with
Qwen3-4B-Thinking-2507 and the real ALFWorldAgentWithMemory. No
training, no Slime / Ray / Megatron — just transformers inference
and the agent loop:
env.reset() -> agent.update_from_env -> model.generate
-> agent.update_from_model (parse tool_call)
-> if memory: agent.execute_memory_tool
else: env.step(wrap_action(tc))
-> agent.update_from_env -> ...
Supports all four compression modes via --mode {lossless_db, lossy,
rag, graph_db}. The graph_db path is the cold-start validation
docs/graph_memory.md "What still needs doing" Accenture#1 calls for: confirm
a base model can emit well-formed entity/relations on a real
trajectory before sinking compute into RL training.
Both scripts are stage-structured: each stage prints OK/FAIL
independently, so a partial failure still tells you which layer broke.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps filelock from 3.20.2 to 3.20.3.
Release notes
Sourced from filelock's releases.
Changelog
Sourced from filelock's changelog.
... (truncated)
Commits
41b42ddFix TOCTOU symlink vulnerability in SoftFileLock (#465)Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting
@dependabot rebase.Dependabot commands and options
You can trigger Dependabot actions by commenting on this PR:
@dependabot rebasewill rebase this PR@dependabot recreatewill recreate this PR, overwriting any edits that have been made to it@dependabot show <dependency name> ignore conditionswill show all of the ignore conditions of the specified dependency@dependabot ignore this major versionwill close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)@dependabot ignore this minor versionwill close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)@dependabot ignore this dependencywill close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)You can disable automated security fix PRs for this repo from the Security Alerts page.