Skip to content

feat(iflow): add support for glm-5, minimax-m2.5, qwen3-32b, tstars2.0, iflow-rome-30ba3b #130

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
MasuRii wants to merge 3 commits into
base: dev
Choose a base branch
from
Open

feat(iflow): add support for glm-5, minimax-m2.5, qwen3-32b, tstars2.0, iflow-rome-30ba3b #130

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
87b894b
feat(iflow): add support for glm-5, minimax-m2.5, qwen3-32b, tstars2.…
MasuRii Feb 15, 2026
d55b566
fix(iflow): address PR review feedback
MasuRii Feb 19, 2026
42a5cf2
fix(iflow): use specific prefixes for reasoning preservation models
MasuRii Feb 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
51 changes: 32 additions & 19 deletions src/rotator_library/providers/iflow_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,27 +29,32 @@

# Model list can be expanded as iFlow supports more models
HARDCODED_MODELS = [
"deepseek-r1",
"deepseek-v3",
"deepseek-v3.1",
"deepseek-v3.2", # seems to not work, but should. Use -chat/-reasoner variants instead
"deepseek-v3.2-chat",
"deepseek-v3.2-reasoner",
"glm-4.6",
"glm-4.7",
"minimax-m2",
"minimax-m2.1",
"qwen3-coder-plus",
"glm-5",
"iflow-rome-30ba3b",
"kimi-k2",
"kimi-k2-0905",
"kimi-k2-thinking", # Seems to not work, but should
"kimi-k2.5", # Seems to not work, but should
"minimax-m2",
"minimax-m2.1",
"minimax-m2.5",
"qwen3-32b",
"qwen3-235b",
"qwen3-235b-a22b-instruct",
"qwen3-235b-a22b-thinking-2507",
"qwen3-coder-plus",
"qwen3-max",
"qwen3-max-preview",
"qwen3-235b-a22b-thinking-2507",
"deepseek-v3.2-reasoner",
"deepseek-v3.2-chat",
"deepseek-v3.2", # seems to not work, but should. Use above variants instead
"deepseek-v3.1",
"deepseek-v3",
"deepseek-r1",
"qwen3-vl-plus",
"qwen3-235b-a22b-instruct",
"qwen3-235b",
"tstars2.0",
]

# OpenAI-compatible parameters supported by iFlow API
Expand Down Expand Up @@ -83,20 +88,30 @@
ENABLE_THINKING_MODELS = {
"glm-4.6",
"glm-4.7",
"glm-5",
"qwen3-max-preview",
"qwen3-32b",
"deepseek-v3.2",
"deepseek-v3.1",
}

# GLM models need additional clear_thinking=false when thinking is enabled
GLM_MODELS = {"glm-4.6", "glm-4.7"}
GLM_MODELS = {"glm-4.6", "glm-4.7", "glm-5"}

# Models using reasoning_split (boolean) instead of enable_thinking
REASONING_SPLIT_MODELS = {"minimax-m2", "minimax-m2.1"}
REASONING_SPLIT_MODELS = {"minimax-m2", "minimax-m2.1", "minimax-m2.5"}

# Models that benefit from reasoning_content preservation in message history
# (for multi-turn conversations)
REASONING_PRESERVATION_MODELS_PREFIXES = ("glm-4", "minimax-m2")
REASONING_PRESERVATION_MODELS_PREFIXES = (
"glm-4",
"glm-5",
"minimax-m2",
"tstars",
"qwen3-32b",
"deepseek-v3.1",
"deepseek-v3.2",
)

# Cache file path for reasoning content preservation
_REASONING_CACHE_FILE = (
Expand Down Expand Up @@ -420,7 +435,7 @@ def _inject_reasoning_content(
"""
Inject cached reasoning_content into assistant messages.

Only for models that benefit from reasoning preservation (GLM-4.x, MiniMax-M2.x).
Only for models that benefit from reasoning preservation.
This is helpful for multi-turn conversations where the model may benefit
from seeing its previous reasoning to maintain coherent thought chains.

Expand Down Expand Up @@ -1022,9 +1037,7 @@ async def stream_handler(response_stream, attempt=1):
else:
if not error_text:
content_type = response.headers.get("content-type", "")
error_text = (
f"(empty response body, content-type={content_type})"
)
error_text = f"(empty response body, content-type={content_type})"
error_msg = (
f"iFlow HTTP {response.status_code} error: {error_text}"
)
Expand Down