[M3gA-Mind]

Summary

What was wrong. response_to_model_response had two gaps.

• Zero-argument tool calls. Native tool-call arguments were parsed with a raw serde_json::from_str. Several providers send a zero-argument call as arguments: "", which is not valid JSON, so the call came back marked invalid.
• Finish reason. The finish reason was always inferred: tool_calls when there were calls, otherwise stop. ChatResponse had no way to carry the provider's reason, so a reply cut off by the output limit was reported as a clean stop, and a caller could not tell to continue or retry.

Fix.

• Empty or whitespace-only arguments are treated as {} before parsing. Other malformed arguments are still kept as invalid calls, unchanged.
• ChatResponse gains finish_reason: Option<String>. When it is set, the conversion keeps it (passed through canonical_finish_reason, so max_tokens/max_output_tokens become length). When it is absent, the old stop/tool_calls inference is used.

Related issue

None.

API or behavior changes

• New public field ChatResponse::finish_reason (Default sets it to None). This is breaking for downstream code that builds ChatResponse with an exhaustive struct literal; such code must add finish_reason or use ..Default::default(). Noted in the changelog. No version bump in this PR.
• A native tool call with empty or blank arguments now converts to a valid call with {}.
• A reported finish reason is kept instead of being replaced with stop/tool_calls.

Validation

• N/A: cargo fmt --all -- --check. I ran rustfmt (edition 2024) on the touched files only; CI runs the full check.
• N/A: cargo clippy --all-targets --all-features -- -D warnings. Not run locally; CI runs it.
• N/A: cargo build --all-targets --all-features. Not run locally; CI runs it.
• N/A: cargo test --all-features. I ran the targeted tests below; CI runs the full suite.

Tests

New response_conversion_tests.rs with two tests.

• empty_or_blank_arguments_mean_an_empty_object
  • Before the fix: FAILED, "" marked invalid
  • After the fix: PASSED.
• reported_finish_reason_is_kept. It also checks that a response with no reported reason still infers tool_calls.
  • Before the fix (field added, conversion unchanged): FAILED, left: Some("stop"), right: Some("length")
  • After the fix: PASSED.

Also run after the fix: cargo test -p tinyinference-llm --lib response (61 passed) and --lib usage::meta (15 passed), which cover the existing conversion tests and the updated ChatResponse literal.

Documentation

Changelog entry under Unreleased. The new field has rustdoc.

Checklist

• The change is focused on one logical change
• No new #[allow(...)], #[ignore], or relaxed lints
• No secrets, tokens, or .env contents in the diff or the description

[tinysweeper]

Tiny Sweeper review

Reviewed head: d1e131a364248aad87f3ea69fa1a16ab6889a4f6

Verdict: Ready for maintainer review · Priority: none
Coverage: Reviewed 5 of 5 files.
Pull request purpose (title): fix(model): keep the…

Diagnostics and supporting details

Full summary

Pull request purpose (title): fix(model): keep the finish reason and accept empty tool arguments in response conversion. Tiny Sweeper completed its review; deterministic results follow.

Review snapshot

Change surface	Files	Review signal	Count
Production	2	Active findings	0
Tests	2	Noted findings	0
Documentation	1	Resolved findings	0
Configuration	0	Pending checks/questions	0

Completeness: Complete
Test assessment: Test coverage is assessed from changed tests and lane evidence; execution is not claimed without trusted check data.

What changed

`ChatResponse` gains a public `finish_reason: Option<String>` field with rustdoc, defaulting to `None`, plus a new `#[path = "response_conversion_tests.rs"]` test module in the conversion file. Conversion now normalizes empty/blank `arguments` to `{}` before parsing, and replaces the fixed `stop`/`tool_calls` assignment with a match that prefers the reported reason via `canonical_finish_reason` and falls back to the existing inference. A changelog entry under Unreleased documents the behavior and the breaking-field requirement. The existing `ChatResponse` literal in `response_tests.rs` is updated with `finish_reason: None`.

Features

• Modified — Empty or blank native tool arguments parse as an empty object: A zero-argument tool call delivered as `arguments: ""` or whitespace-only no longer becomes an invalid call; it converts to a valid call with `{}`. Other malformed arguments still produce invalid calls with the original argument text and parse error preserved. (crates/tinyinference-llm/src/model/response_conversion.rs pub fn response_to_model_response()
• Added — Provider finish reason carried on ChatResponse and preserved in conversion: `ChatResponse::finish_reason` lets the adapter record the provider's reason; when set it passes through `canonical_finish_reason` (e.g. `max_tokens` → `length`), so a length-truncated reply is no longer reported as a clean `stop` and callers can decide to continue or retry. When absent, the prior `stop`/`tool_calls` inference applies. Adding the field is breaking for exhaustive struct literals. (crates/tinyinference-llm/src/providers/response.rs pub struct ChatResponse<U = crate::usage::Usage, T = crate::tool::ToolCall> {, crates/tinyinference-llm/src/providers/response.rs impl<U, T> Default for ChatResponse<U, T> {, crates/tinyinference-llm/src/model/response_conversion.rs pub fn response_to_model_response(, CHANGELOG.md)

Tests

• existing_test_update — The `ChatResponse` literal in the existing clone-preservation test gains `finish_reason: None` to satisfy the new field.: Keeps the existing test compiling against the new struct shape; the change is a literal initialization only. Not evidence of execution. (crates/tinyinference-llm/src/providers/response_tests.rs fn raw_call_identity_arguments_signature_and_exact_billing_survive_a_clone() {)

Findings

No active actionable findings.

Before merge

None.

Agent review details

critique

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: Reviewed 5 files; 0 findings. _The code index is behind this pull request (indexed at `50e6d1ce19bd`), so retrieved context may be out of date._

security

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The security lane reviewed 4 code files with 0 findings and explicitly excluded `CHANGELOG.md` as prose data.
• Lane summary: Reviewed 4 files; 0 findings. 1 file was not security-reviewed: CHANGELOG.md (prose or tabular data). _The code index is behind this pull request (indexed at `50e6d1ce19bd`), so retrieved context may be out of date._

tests

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The two behavioural changes each have a dedicated test that would fail on regression: `empty_or_blank_arguments_mean_an_empty_object` pins the blank/empty `arguments` handling in the new branch, and `reported_finish_reason_is_kept` pins both the pass-through through `canonical_finish_reason` (`max_tokens` → `length`) and the unreported fallback. One gap remains: the untested `None if tool_calls.is_empty()` arm of the new finish-reason match, though the tool-call path of the fallback is covered and the removed `stop` inference is a pre-existing, unchanged behaviour. No defects otherwise. _The code index is behind this pull request (indexed at `50e6d1ce19bd`), so retrieved context may be out of date._

commits

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The commits lane reported no sensitive content in committed files.
• Lane summary: No sensitive content in committed files. _The code index is behind this pull request (indexed at `50e6d1ce19bd`), so retrieved context may be out of date._

description

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The change does what the description says: `response_to_model_response` now maps empty or whitespace-only native tool arguments to `{}` before parsing, and keeps the provider's finish reason (canonicalized) via the new `ChatResponse::finish_reason` field, falling back to the old `stop`/`tool_calls` inference. Remaining `ChatResponse` literals in the crate use `..Default::default()` or are updated, the new test file follows the repository's `*_tests.rs` convention, and the changelog entry covers the breaking-field note. The description is accurate and looks sound to merge. _The code index is behind this pull request (indexed at `50e6d1ce19bd`), so retrieved context may be out of date._

Evidence and run details

• Models: ladder/vectors-oai3, deepseek-v4.1-flash
• Spend: $0.000659
• Tokens: 236498 input · 4813 output · 48128 cached · 259 embedding
• Updated: 2026-10-22T23:01:17Z

Head	State	Pass summary
d1e131a36424	ready for maintainer review	0 active finding(s), 0 resolved finding(s) (at 2026-10-22T23:01:17Z)

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 6fe24a31-62b9-4f67-b41d-c73b36a33a6e

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[tinysweeper]

tinysweeper found nothing blocking. Approving.

Diagnostics

Total spend: $0.0007