cd /news/large-language-models/deepseek-v4-flash-0731-tool-call-rep… · home topics large-language-models article
[ARTICLE · art-126440] src=gist.github.com ↗ pub= topic=large-language-models verified=true sentiment=↓ negative

DeepSeek-V4-Flash-0731 tool-call repetition collapse: portable reproducer (3 stacks incl. Ollama cloud 6/6)

A developer at Petabit Scale published a portable reproducer showing that DeepSeek-V4-Flash-0731 collapses into repeating the short action-announcement "Run." instead of emitting tool calls during long agentic conversations. The failure was reproduced on three independent serving stacks — Ollama cloud (6/6 trials), Alibaba DashScope international (2/6), and a community SM80 vLLM fork on 8x A100 (4/6) — using a sanitized ~240k-token, 394-message conversation distilled from a production incident on 2026-08-12/13. Synthetic conversations matching the same shape produced 0/22 loops, suggesting the attractor depends on realistic content statistics; the issue was reported to DeepSeek on 2026-08-14.

read1 min views5 publishedAug 14, 2026

One fixed, fully sanitized ~240k-token agentic conversation that makes DeepSeek-V4-Flash-0731 collapse into repeating a short action-announcement ("Run.") instead of emitting a tool call. Reproduced 2026-08-14 on three serving stacks that share no code or hardware:

Serving stack Trials Strict loops Notes
Ollama cloud, deepseek-v4-flash:0731-cloud 6 6/6 61-97x "Run.", all rolls hit the 2,048-token cap
Alibaba DashScope intl, deepseek-v4-flash 6 2/6 remaining 4 rolls emitted 6-19x "Run." before the tool call
Community SM80 vLLM (haosdent fork), 8x A100 6 4/6 where we first hit it in production

Strict loop metric (implemented in the script): a normalized sentence-segment occurring >=20 times in one completion, or >=24 consecutive segments of <=5 words. Sampling: temperature 1.0, top_p 0.95, no penalties, tool_choice=auto.

python3 cross_check.py <openai_compatible_base_url> <api_key> <model> <tag> 6

Run from the directory containing dsv4-repro-payload-sanitized.json.

Distilled from a production incident (Petabit Scale, 2026-08-12/13). The conversation is real (394 messages, 210 tool results, 51 declared tools) with every identifier neutralized; sanitization did not remove the trigger. Notably, synthetic conversations matching its shape (depth, tool count, structure, mid-chain ending, repeated-history priming) produced 0/22 loops - the attractor appears to require realistic content statistics. Related: ggml-org/llama.cpp#26694 (Flash-0731 GGUF/Metal), QwenLM/qwen-code#4695 (V4-Pro on DashScope, incl. a dose-response showing in-context self-reinforcement). Reported to DeepSeek (service@deepseek.com) with full receipts on 2026-08-14. Contact: ras@petabitscale.com

── more in #large-language-models 4 stories · sorted by recency
── more on @deepseek 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/deepseek-v4-flash-07…] indexed:0 read:1min 2026-08-14 ·