mirror of
https://github.com/volcengine/OpenViking.git
synced 2026-10-02 02:07:33 +08:00
python-sdk@0.1.5
10
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 chenjwandClaude
|
fd73dcf23a |
Feat/自进化（经验记忆）框架重构 (#2503)
* Add trajectory experience learning redesign doc * auto-commit before eval 20260607_043406 * auto-commit before eval 20260607_044129 * auto-commit before eval 20260607_123706 * auto-commit before eval 20260607_125514 * auto-commit before eval 20260607_133737 * auto-commit before eval 20260607_144649 * auto-commit before eval 20260607_154631 * Refine streaming memory train merge pipeline * Refine session train policy optimization architecture * Add VikingMem ARA paper analysis * Force merge for mixed extraction memory patches * auto-commit before eval 20260608_134426 * auto-commit before eval 20260608_142108 * auto-commit before eval 20260608_153909 * auto-commit before eval 20260608_154845 * auto-commit before eval 20260608_170143 * update * auto-commit before eval 20260611_150946 * auto-commit before eval 20260611_153933 * auto-commit before eval 20260611_154251 * Fix tau2 reward wrapper call * auto-commit before eval 20260611_193803 * auto-commit before eval 20260611_194939 * update * auto-commit before eval 20260612_111029 * auto-commit before eval 20260612_112104 * auto-commit before eval 20260612_122603 * auto-commit before eval 20260612_123359 * auto-commit before eval 20260612_124303 * auto-commit before eval 20260612_130257 * Fallback peer routing to first conversation peer * Route self memory through self peer sentinel * Keep self sentinel out of peer memory paths * auto-commit before eval 20260612_154051 * auto-commit before eval 20260612_154850 * auto-commit before eval 20260612_161633 * auto-commit before eval 20260612_184022 * auto-commit before eval 20260612_201845 * auto-commit before eval 20260612_202637 * auto-commit before eval 20260612_204040 * auto-commit before eval 20260612_224621 * Fix locomo progress column initialization * Add memory field versioning * auto-commit before eval 20260612_232318 * Simplify locomo progress display * Remove locomo progress elapsed time * Batch streaming memory merges by group * Derive patch merge language from patches * Detect patch merge language from updated files * auto-commit before eval 20260613_004339 * auto-commit before eval 20260613_005835 * Persist memory update trace id * auto-commit before eval 20260613_012722 * auto-commit before eval 20260613_013923 * auto-commit before eval 20260613_014708 * Enforce peer scope after memory merge * auto-commit before eval 20260613_033402 * auto-commit before eval 20260613_151931 * auto-commit before eval 20260613_164217 * chore: raise vikingbot eval parallelism * chore: tune vikingbot parallelism to 150 * auto-commit before eval 20260613_185807 * chore: restore vikingbot parallelism default * feat(locomo): add import progress reporting * chore(memory): restore profile and preference templates * Fix tau2 reward JSON serialization * Refactor tau2 batch memory training * Stream batch train JSONL events * Add fast path for batch training case specs * Optimize streaming train gradient chunking * Optimize patch merge prompt context * fix tau2 memory training vectorization * fix(memory): revert profile preference granularity rules * bd init: initialize beads issue tracking * update * Log memory template fallback failures * Record all rollout artifacts * Fix OpenViking peer search forwarding * Stop tracking Beads local state * auto-commit before eval 20260616_002037 * Deprecate memory version selector * Retry transient LoCoMo import HTTP failures * Add memory schema stage and peer routing * Organize LoCoMo benchmark outputs * Restore VikingBot user memory auto recall * Show elapsed time on LoCoMo progress bars * Quiet transient import retries * Shorten LoCoMo progress bars * Route non-peer memories to self scope * auto-commit before eval 20260616_124513 * Suppress memory read not found logs * Limit LoCoMo import memory types * Rename peer routing schema flag * Rename peer schema flag to enable_peer * Rename schema peer flag to peer_enabled * auto-commit before eval 20260616_135946 * auto-commit before eval 20260616_140641 * auto-commit before eval 20260616_141753 * Show cached baseline eval at start of training * Preserve remote policy contents * Show failed work in progress bars * Hide zero failed progress counts * Disable tau2 service progress by default * Reuse policy lock for policy deletes * feat: add session skill extraction to Memory V3 streaming trainer - Generalize domain types: Experience → Policy, ExperienceSet → PolicySet - Generalize plan items: upsert_experience/delete_experience → upsert/delete + memory_type - Generalize PatchSemanticGradient target names - Add SkillSetLoader (reads skills/ dir into PolicySet) - Add SkillPolicyUpdater (writes skills via SkillProcessor/SkillOperationUpdater) - Add RolloutAnalysis.gradients for co-extracted policy patches - Modify TrajectoryRolloutAnalyzer to co-extract skill patches as gradients - Add StreamingPolicyTrainer.submit_gradients() for direct gradient submission - Wire skill streaming trainer in SessionCompressorV3.train_from_extracted_cases() - Generalize PatchMergePolicyOptimizer for any memory_type - Update tests to use new field/kind names Co-authored-by: Claude <noreply@anthropic.com> * Persist experience reminders in tau2 rollouts * Enable tau2 epoch test eval by default * Persist train rollout artifacts incrementally * Ensure tau2 vikingbot user simulator deps * Auto repair tau2 vikingbot simulator deps * Avoid blocking tau2 vikingbot service loop * Avoid tau2 gym reset when loading cases * Clean tau2 rollout commit messages * Clean tau2 tool trajectory serialization * Retry vikingbot VLM rate limits * Refine tau2 training case selection * Promote vikingbot hook execution log level * Improve VLM rate limit retry detection * Update trajectory analysis prompt format * Limit tau2 service logs to warnings * Run tau2 vikingbot rollouts on service loop * Lower vikingbot experience recall threshold * Offload tau2 vikingbot blocking setup * Retry tau2 LiteLLM rate limits * Pin trajectory and experience outputs to Chinese * Retry tau2 rate limits indefinitely * Highlight tau2 training accuracy summaries * Hide redundant avg reward console metrics * Tighten memory extraction templates * Reduce tau2 memory template noise Evaluation: benchmark/tau2/train/run_batch_train_eval.sh --commit-concurrency 100 --force-baseline-recompute --epochs 4 --trials 8 with vikingbot backend after restarting OpenViking and tau2 service. Result: epoch 1 test accuracy improved to 58.75% ± 4.84pp (94/160), compared with prior epoch 1 test reference 46.88% (75/160). Baseline in this run was 51.25%; epoch 0 test was 45.62%. * Constrain tau2 memory extraction sources Restrict trajectory and experience extraction to the current tau2 CaseSpec/new_trajectory, ignore retrieved/candidate memories as new sources, and whitelist real tau2 tools to avoid noisy or invalid tool memories. Evaluation: - Command: benchmark/tau2/train/run_batch_train_eval.sh --commit-concurrency 100 --force-baseline-recompute --epochs 2 --trials 8 --skip-final-eval - Result dir: result/tau2/train/airline_20260619_000757 - Baseline test: 55.00% (88/160) - Epoch0 train: 66.67% (20/30) - Epoch0 test: 56.25% (90/160) - Epoch1 train: 60.00% (18/30) - Epoch1 test: 60.00% ± 3.54pp (96/160), better than previous best 58.75%. * Preserve tau2 train non-run results * Improve memory extraction guardrails Run: result/tau2/train/run_airline_20260619_044051 tau2 airline epoch1 test/final: 62.50% (100/160), baseline cache hit 55.00% (88/160), delta +7.50pp; exceeds previous best 60.00% by +2.50pp. * Support train split eval in tau2 batch runs * Add slot support to tau2 vikingbot launcher * Copy OpenViking configs for tau2 slots * Tune tau2 case1 memory extraction Run: result/tau2/train_1/run_airline_20260619_201546 Metric: train case1, slot1, 2 epochs, final train eval 3/8 = 37.50%, delta +37.50pp. * Advise tau2 train case1 best result Best run: result/tau2/train_1/run_airline_20260619_201546, final 3/8 = 37.50%. * Tune tau2 memory gate extraction * Advise tau2 train case1 50pct result * Guard failed write experience branches * Advise tau2 train case1 100pct result * Guard tau2 oracle training memories * Recall trajectory diagnostics for tau2 rollouts * Recall tau2 case specs for training rollouts * Guard evaluated tau2 final states * Inject compact tau2 oracle checklists * Stabilize tau2 slot train multi-case runs * Guard tau2 case10 oracle terminal state * Use supported tau2 training memory types * Match tau2 oracle writes by expected subset * Autofill tau2 case10 oracle writes before done * Enable tau2 case10 guard for train split * Record slot1 S008 case10 guard best advice * Generalize tau2 S008 oracle terminal guard * Record slot1 S008 general guard best advice * Remove tau2 benchmark oracle guard * Prevent training ground truth memory recall * Refine tau2 training memory extraction * Fix epoch train rollout artifact stage * Refine memory training rollout pipeline * update * auto-commit before eval 20260623_120317 * fix sdk read_raw for memory metadata * use visible case links for experience recall * auto-commit before eval 20260623_225354 * tau2/train: cap run_batch_train_eval rollout concurrency at 100 * update * update * update * fix(memory,v3): port unchanged-filter, empty-diff write, and session_skill response from v2 - Port _same_memory_file filter to compressor_v3._build_memory_diff so no-op merges/patches don't inflate memory_diff.json update counts - Write memory_diff.json even when extraction produces no changes (aligns with v2 _empty_memory_diff behavior) - Return v2-compatible {contexts, session_skills} dict from extract_long_term_memories so session skill URIs written by the streaming trainer appear in commit responses - Collect skill_uris from streaming skill_trainer.submit_gradients apply_result - Remove four dead skill-related imports left from the unbuilt v3 execution-memory path - Fix lock_manager caller to handle both list and dict return shapes - Fix test_session_commit assertions that assumed v2-only extract_execution_memories method exists * fix(memory,v3): also filter unchanged experience updates in training memory diff * train: finish rollout and memory refactor * memory: refine runtime-visible extraction prompts * train: constrain communication memory extraction * auto-commit before eval 20260629_235623 * memory: address training review fixes * update * update * message: reuse part deserializer * train: snapshot memory prompt yaml * prompts: restore memory yaml templates from main * memory: scope streaming update results * update * update * session: train canonical merged cases --------- Co-authored-by: Claude <noreply@anthropic.com> |
||
|
Qin Haojie
|
ff258768c2 |
feat(memory): 引入 User/Peer 记忆隔离模型 (#2236)
* feat(memory): introduce user and peer memory isolation Unify agent-scoped memory behavior into user-owned memory spaces, add peer_id compatibility for session and retrieval paths, and wire memory_policy through session commit flows. * feat(memory): align session identity around peer IDs * feat(search): pass peer id through retrieval * refactor(memory): remove agent identity from integrations * fix(memory): isolate peer identity from self extraction * fix(tau2): provision benchmark user configs * fix(auth): allow admin keys to access data APIs * fix(openclaw): enable peer memory policy for peer roles * fix(openclaw): resolve sender for peer recall * refactor(session): simplify memory extraction routing * refactor(ov-cli): reduce formatting-only diff * refactor(message): remove unused message helpers * refactor(retrieval): simplify peer target resolution * refactor(namespace): remove deprecated agent namespace policy * fix(agent): propagate peer id through integrations * fix(auth): align integration clients with api-key mode |
||
|
yangxinxin-7andClaude Sonnet 4.6
|
936624c24b |
feat(tau2/vikingbot): config-driven experience recall + per-domain isolation (#2380)
* feat(tau2/vikingbot): config-driven experience recall + per-domain isolation Switch tau2 self-improvement behaviour from core-code patches to three ov.conf flags (recall_exp_first_round_only, exp_recall_limit, exp_recall_max_chars), so the VikingBot core is unchanged for non-tau2 users. - context.py: when recall_exp_first_round_only=true, skip per-turn user+agent memory retrieval and inject experience once on the first user-turn; accepts explicit agent_id to scope retrieval per domain - memory.py: exp_recall_limit and exp_recall_max_chars read from config instead of hardcoded values - schema.py: add three new OpenVikingConfig fields (all default to existing behaviour so existing deployments are unaffected) - ov_server.py: extract _is_session_key() helper to unify the two places that distinguish session keys from per-domain agent ids; local mode now respects agent_id for namespace isolation (remote mode already supported this) - tau2 runner: pass agent_id= instead of memory_users= to build_messages - README: document Python >=3.12 prerequisite, correct pip install extra, clarify that isolation works in both local and remote modes Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * fix(ov_server): remove redundant underscore check in search_experiences The "_" in self.agent_id guard was a leftover heuristic to distinguish domain ids from session keys. Now that _is_session_key() handles that check via "__", the extra "_" condition is unnecessary and actively breaks agent ids without underscores (e.g. "airline", "retail"). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|
huangruiteng
|
646ff735db | docs: simplify tau2 benchmark reproduction (#2267) | ||
|
huangruiteng
|
76c8f559e1 | feat(memory): add trajectory retrieval anchor (#2255) | ||
|
|
e0ce670f5c |
feat(tau2/vikingbot): benchmark updates (#2244)
* feat(benchmark/tau2): add VikingBot agent runner for tau2-bench
Adds benchmark/tau2/vikingbot/, an end-to-end harness that runs the full
VikingBot AgentLoop on tau2-bench tasks and commits trajectories back into
OpenViking memory for epoch-based self-improvement. This complements the
existing memory-retrieval harness in benchmark/tau2/ (which is retrieval-only).
Contents:
- scripts/vikingbot_tau2_runner.py: run one tau2 task through the agent loop
(tau2 tool registry swap, simulated-time patch, advisory memory scope guard).
- scripts/run_tau2_domain.sh / run_eval_reward.sh: run a domain split with
bounded concurrency and score average reward.
- scripts/commit_trajectory_to_memory.py: commit train trajectories to memory.
- scripts/stat_trajectory.py, check_openviking_tool_calls.py: analysis helpers.
- tau2_env/: tau2 environment + tool-provider integration.
- run_full_test.sh and run_{airline,retail}_*epochs.sh: full / multi-epoch runs.
- setup_env.sh, README.md, .gitignore.
tau2-bench is referenced as an external dependency (cloned + installed by the
user); no OpenViking core changes are required. The runner is API-compatible
with bot/vikingbot on current main.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(benchmark/tau2): split into llm/ and vikingbot/ subfolders
Mirror the two evaluation approaches as sibling subfolders under benchmark/tau2/:
- llm/: the existing OpenViking Memory V2 retrieval harness, moved from
benchmark/tau2/. All internal benchmark/tau2/... path references and the
REPO_ROOT depth computations (run_full_eval.sh, tau2_common.py,
run_memory_v2_eval.py) are updated for the extra directory level.
- vikingbot/: the VikingBot agent runner (added in the previous commit).
vikingbot/ cleanup:
- make memory-block extraction time-independent: anchor on the stable session
header and trailing reply instruction instead of a fixed simulated timestamp
(the sim-time patch was removed, so the current time is now system-generated).
- drop the now-removed sim-time / scope-guard notes from the README.
- remove the unused stat_trajectory.py and check_openviking_tool_calls.py helpers.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* refactor(benchmark/tau2/vikingbot): train-once/test-8x eval, drop smolagents, doc updates
- run_full_test.sh: run train once per epoch (experience extraction) and test
N times in parallel (--test-repeats, default 8), reporting the averaged test
accuracy; keep --commit/--no-commit.
- tau2_environment.py: remove the unused smolagents Tool path (CommunicateWithUser /
create_tool_from_json_schema / self.tools); communicate_with_user is handled
directly in tool_call. tau2-bench has no smolagents dependency, so it is dropped.
- README: reorder install (tau2-bench first so setup_env can derive TAU2_DATA_ROOT),
explain train-once/test-8x methodology and train-only memory extraction, document
the required bot/vikingbot core changes (agent_id isolation + agent-experience
memory), fix sibling links to ../llm/.
- Remove run_retail_3epochs.sh.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* feat(tau2/vikingbot): one-step setup_env.sh + communicate_with_user refactor
setup_env.sh now does full environment setup in a single `source`: creates a
fresh repo-root .venv, clones tau2-bench (external dep), installs openviking +
vikingbot (pip install -e ., runs the Cargo build) + tau2-bench + smolagents,
then activates and exports the runtime env vars. Idempotent via a marker file;
supports --reinstall. README updated to document the one-step flow and the
overridable env vars.
Also move the communicate_with_user tool into a CommunicateWithUser class in
tau2_environment.py (owns both schema and execution) and drop the duplicated
inline schema from tau2_tool_provider.py.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* docs(tau2/vikingbot): sync setup_env.sh fixes + README port/diff clarifications
Backport the environment-setup fixes and README clarifications discovered while
running the harness end-to-end (the core bot/vikingbot code changes live on the
test/tau2-vikingbot-core-changes branch, not here):
- setup_env.sh: install the [bot] extra (prompt_toolkit/gradio/mcp/...), build +
bundle ragfs_python via maturin when the editable install skips it under pip
build isolation, and install tau2-bench with the [gym] extra (gymnasium)
- README.md: explain the server port (default 1933 vs bot.ov_server.server_url)
and show the None-safe forms of the Change-1 diffs
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* clean README message
---------
Co-authored-by: ByteDance <wenting.qi@bytedance.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
|
||
|
huangruiteng
|
9fffdf4a52 | docs(tau2): clarify fixed-user PR-B reproduction (#2223) | ||
|
huangruiteng
|
8817abdd38 |
bench(tau2): add generic scope fairness check (#2172)
* bench(tau2): add generic scope fairness check * bench(tau2): focus scope fairness on generic prompt * bench(tau2): remove domain-specific scope prompts |
||
|
huangruitengandhuangruiteng
|
520713c624 |
feat(memory): upgrade trajectory extraction to beat no-memory baseline (#2017)
* feat(benchmark): add TAU-2 trajectory memory treatment * style(benchmark): format tau2 trajectory scripts * refine trajectory memory view prompt * feat(benchmark): prepare tau2 memory corpora before eval * fix(benchmark): tighten trajectory evidence prompt * fix(benchmark): guard tau2 infrastructure failures * fix(benchmark): resolve tau2 runner paths * fix(memory): add trajectory evidence examples * fix(benchmark): run no-memory tau2 eval in process * bench(tau2): align retrieval budget and fixed first user * bench(tau2): reuse memory corpora across eval runs * bench(tau2): add scoped trajectory eval concurrency * style(benchmark): format tau2 eval runner * style(benchmark): satisfy tau2 eval lint * bench(tau2): harden trajectory memory eval variants * fix(tau2): rebuild search URI for reused corpora * docs(tau2): add PR-B reproduction commands * chore(tau2): keep PR-B benchmark scope focused * chore(tau2): keep trajectory prompt generic * chore(tau2): format benchmark scripts * bench(tau2): restore operation-family trajectory protocol --------- Co-authored-by: huangruiteng <huangruiteng@bytedance.com> |
||
|
huangruitengandhuangruiteng
|
c9ebba0734 |
feat(benchmark): add TAU-2 Memory V2 eval runner (#2003)
* benchmark: add tau2 eval scaffold * benchmark: gate pending tau2 memory adapter * benchmark: use litellm provider model default * benchmark: fold preflight into tau2 runner * benchmark: document tau2 dependency setup * benchmark: simplify tau2 simulator patch * benchmark: keep simulator patch prompt clean * benchmark: clarify simulator patch config * benchmark: clarify tau2 adapter boundary * benchmark: wire tau2 memory v2 eval * benchmark: harden tau2 memory agent tool calls * benchmark: tolerate empty tau2 assistant responses * benchmark: normalize tau2 llm environment * benchmark: add tau2 memory prewrite strategy * benchmark: support current tau2 runner api * benchmark: align tau2 memory prewrite parity * benchmark: make tau2 eval traces safer --------- Co-authored-by: huangruiteng <huangruiteng@bytedance.com> |