problem · Revision 1
Cause (Documented platform behavior): The MPS allocator enforces an upper bound (high watermark ratio); crossing it raises this error. Fix status: documented_behavior Other error fragments: - Use PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0 to disable upper limit for memory allocations (
problem · Revision 1
Cause (Documented platform behavior): The Temporal runner only routes model calls through activities for async Runner.run/run_streamed; run_sync is rejected in workflows. Fix status: documented_behavior Evidence (public sources, summarized; not reproduced by this contributor): -
problem · Revision 1
Cause (Documented platform behavior): RedisVL checks the server version and search/searchlight module versions before creating SVS-VAMANA fields. Fix status: documented_behavior Other error fragments: - SVS-VAMANA only supports FLOAT16 and FLOAT32 datatypes. Evidence (public sour
problem · Revision 1
Cause (Documented platform behavior): FlashInfer compiles kernels at runtime and needs the CUDA toolkit and ninja. Fix status: documented_behavior Other error fragments: - FlashInfer failed to JIT-compile: ninja (build tool) not found. Evidence (public sources, summarized; not re
problem · Revision 1
Cause (Documented platform behavior): Verifier tracks active messages/tool calls by ID and enforces START -> CONTENT/ARGS -> END ordering. Fix status: documented_behavior Other error fragments: - Cannot send 'TOOL_CALL_ARGS' event: No active tool call found with ID - A text
problem · Revision 1
Cause (Documented platform behavior): When PYTORCH_ENABLE_MPS_FALLBACK is unset or "0", unsupported MPS ops dispatch to an error fallback; otherwise to a CPU fallback. Fix status: documented_behavior Unknowns: - Whether setting the variable via os.environ after importing torch ta
problem · Revision 1
Cause (Documented platform behavior): The worker detects workflow activations that do not yield within the timeout and fails the task with the thread stack trace. Fix status: documented_behavior Evidence (public sources, summarized; not reproduced by this contributor): - https://
problem · Revision 1
Cause (Documented platform behavior): Extensions compare the existing index schema with the one implied by the current configuration and refuse to silently reuse a mismatched index. Fix status: documented_behavior Limitations: - Whether overwrite=True also deletes existing docume
problem · Revision 1
Cause (Documented platform behavior): The allocator could not find a contiguous block; the message distinguishes memory used by other processes, memory allocated by PyTorch, and cached-but-unallocated memory (fragmentation). Fix status: documented_behavior Other error fragments:
problem · Revision 1
Cause (Documented platform behavior): The verifier requires every opened message/tool call/step/span to be closed before RUN_FINISHED. Fix status: documented_behavior Other error fragments: - Cannot send 'RUN_FINISHED' while text messages are still active: - Cannot send 'RUN_FINI
problem · Revision 1
Cause (Documented platform behavior): Redis Search reports an absent index as an ordinary error reply with no error code, and the wording changed across versions. Fix status: documented_behavior Misleading approaches: - Treating any FT.INFO error as "missing": permission errors (
problem · Revision 1
Cause (Documented platform behavior): Verifier tracks run state and rejects events after terminal events. Fix status: documented_behavior Other error fragments: - The run has already errored with 'RUN_ERROR'. No further events can be sent. - Cannot send 'RUN_STARTED' while a
problem · Revision 1
Cause (Documented platform behavior): Policy must match [quota];w=[seconds](;u=request|cents)(;s=segment) with 60 <= window <= 31536000 and positive quota. Fix status: documented_behavior Other error fragments: - Window must be at least 60 seconds - Quota must be a positive numbe
problem · Revision 1
Cause (Documented platform behavior): The client needs a file name to infer MIME type for the upload. Fix status: documented_behavior Evidence (public sources, summarized; not reproduced by this contributor): - https://raw.githubusercontent.com/run-llama/llama_cloud_services/f385
problem · Revision 1
Cause (Documented platform behavior): The client verifier enforces the AG-UI lifecycle: every run must begin with RUN_STARTED. Fix status: documented_behavior Evidence (public sources, summarized; not reproduced by this contributor): - https://raw.githubusercontent.com/ag-ui-prot
problem · Revision 1
Cause (Documented platform behavior): The worker requires Helicone-Auth to contain "Bearer " followed by a key matching Helicone API key patterns. Fix status: documented_behavior Limitations: - HTTP status code for this path not confirmed from the snippet read. Other error
problem · Revision 1
Cause (Documented platform behavior): encoding_for_model maps by exact name then known prefixes; anything else raises KeyError. Fix status: documented_behavior Other error fragments: - to a tokeniser. Please use `tiktoken.get_encoding` to explicitly get the tokeniser you expect.
problem · Revision 1
Cause (Documented platform behavior): Core matches each history event to the next command emitted by replayed workflow code; a mismatch raises a Nondeterminism error. Fix status: documented_behavior Workaround (not a fix): Setting workflow_failure_exception_types to include Nonde
problem · Revision 1
Cause (Documented platform behavior): The client polls job status with backoff and raises a generic Exception when elapsed time exceeds max_timeout; job itself may still complete server-side. Fix status: documented_behavior Misleading approaches: - Catching a specific timeout exc
problem · Revision 1
Cause (Documented platform behavior): tiktoken downloads encoding files over HTTPS on first use and caches them under TIKTOKEN_CACHE_DIR (or DATA_GYM_CACHE_DIR, else <tmp>/data-gym-cache). The cache file name is the SHA-1 hex digest of the blob URL, and the contents are checked