mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-05-08 03:01:47 +00:00
fix(tool-schemas): reactive strip of pattern/format on llama.cpp grammar 400s
MCP servers commonly emit JSON Schema `pattern` (e.g. `\\d{4}-\\d{2}-\\d{2}`
for date-time params) and `format` keywords. llama.cpp's
`json-schema-to-grammar` converter rejects regex escape classes
(\\d/\\w/\\s) and most format values, returning HTTP 400
"parse: error parsing grammar: unknown escape at \\d" — the whole request
fails.
Cloud providers (OpenAI, Anthropic, OpenRouter, Gemini) accept these
keywords fine and use them as prompting hints. Stripping unconditionally
loses useful hints for every cloud user to fix a llama.cpp-only bug.
Approach: classify the llama.cpp grammar-parse 400 in the error
classifier, and on match do a one-shot in-place strip of pattern/format
from `self.tools`, then retry. Follows the existing
`thinking_signature` recovery pattern. Cloud users hit zero overhead;
llama.cpp users pay one failed request per session.
Changes
- agent/error_classifier.py: new `FailoverReason.llama_cpp_grammar_pattern`
+ narrow HTTP-400 branch matching "error parsing grammar",
"json-schema-to-grammar", or "unable to generate parser ... template".
- tools/schema_sanitizer.py: new `strip_pattern_and_format()` helper —
reactive, walks schema nodes, skips property names (search_files.pattern
survives). Returns strip count for logging.
- run_agent.py: new one-shot recovery block in the retry loop. Strips,
logs, continues. Falls through to normal retry if nothing to strip.
- tests: 4 classifier tests (3 variants + 1 non-400 negative), 7 strip
tests including the property-name preservation and idempotency checks.
Co-authored-by: Chris Danis <cdanis@gmail.com>
This commit is contained in:
parent
542e06c789
commit
28f4d6db63
5 changed files with 280 additions and 1 deletions
44
run_agent.py
44
run_agent.py
|
|
@ -11116,6 +11116,7 @@ class AIAgent:
|
|||
thinking_sig_retry_attempted = False
|
||||
image_shrink_retry_attempted = False
|
||||
oauth_1m_beta_retry_attempted = False
|
||||
llama_cpp_grammar_retry_attempted = False
|
||||
has_retried_429 = False
|
||||
restart_with_compressed_messages = False
|
||||
restart_with_length_continuation = False
|
||||
|
|
@ -12206,6 +12207,49 @@ class AIAgent:
|
|||
)
|
||||
continue
|
||||
|
||||
# ── llama.cpp grammar-parse recovery ──────────────────
|
||||
# llama.cpp's ``json-schema-to-grammar`` converter rejects
|
||||
# regex escape classes (``\d``, ``\w``, ``\s``) and most
|
||||
# ``format`` values in tool schemas. MCP servers emit
|
||||
# these routinely for date/phone/email params. Recovery:
|
||||
# strip ``pattern``/``format`` from ``self.tools`` and
|
||||
# retry once. We keep the keywords by default so cloud
|
||||
# providers get the full prompting hints; this branch
|
||||
# fires only for users on llama.cpp's OAI server.
|
||||
if (
|
||||
classified.reason == FailoverReason.llama_cpp_grammar_pattern
|
||||
and not llama_cpp_grammar_retry_attempted
|
||||
):
|
||||
llama_cpp_grammar_retry_attempted = True
|
||||
try:
|
||||
from tools.schema_sanitizer import strip_pattern_and_format
|
||||
_, _stripped = strip_pattern_and_format(self.tools)
|
||||
except Exception as _strip_exc: # pragma: no cover — defensive
|
||||
logging.warning(
|
||||
"%sllama.cpp grammar recovery: strip helper failed: %s",
|
||||
self.log_prefix, _strip_exc,
|
||||
)
|
||||
_stripped = 0
|
||||
if _stripped:
|
||||
self._vprint(
|
||||
f"{self.log_prefix}⚠️ llama.cpp rejected tool schema grammar — "
|
||||
f"stripped {_stripped} pattern/format keyword(s), retrying...",
|
||||
force=True,
|
||||
)
|
||||
logging.warning(
|
||||
"%sllama.cpp grammar recovery: stripped %d "
|
||||
"pattern/format keyword(s) from tool schemas",
|
||||
self.log_prefix, _stripped,
|
||||
)
|
||||
continue
|
||||
# No keywords found to strip — fall through to normal
|
||||
# retry path rather than loop forever on the same error.
|
||||
logging.warning(
|
||||
"%sllama.cpp grammar error but no pattern/format "
|
||||
"keywords to strip — falling through to normal retry",
|
||||
self.log_prefix,
|
||||
)
|
||||
|
||||
retry_count += 1
|
||||
elapsed_time = time.time() - api_start_time
|
||||
self._touch_activity(
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue