部署Minimax2.1后，回复结果混乱，所答非所问

[wurenxiaozu]

Reminder

• I have read the above rules and searched the existing issues.

System Info

root@iv-ydps7tineowh2yorhj7f:/appdata/sana# nvidia-smi
Sat Jan 3 14:26:35 2026
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.161.08 Driver Version: 535.161.08 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA L20 On | 00000000:65:01.0 Off | 0 |
| N/A 34C P0 77W / 350W | 38124MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
| 1 NVIDIA L20 On | 00000000:67:01.0 Off | 0 |
| N/A 27C P8 35W / 350W | 3MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
| 2 NVIDIA L20 On | 00000000:69:01.0 Off | 0 |
| N/A 33C P0 77W / 350W | 17308MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
| 3 NVIDIA L20 On | 00000000:6B:01.0 Off | 0 |
| N/A 27C P8 33W / 350W | 3MiB / 46068MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| 0 N/A N/A 2820822 C python3 13312MiB |
| 0 N/A N/A 2821173 C python3 13312MiB |
| 0 N/A N/A 2821423 C python3 11488MiB |
| 2 N/A N/A 2814512 C /usr/bin/ollama 17302MiB |
+---------------------------------------------------------------------------------------+
root@iv-ydps7tineowh2yorhj7f:/appdata/sana# lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 52 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 90
On-line CPU(s) list: 0-89
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8457C
CPU family: 6
Model: 143
Thread(s) per core: 2
Core(s) per socket: 45
Socket(s): 1
Stepping: 8
BogoMIPS: 5200.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx
fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopolog
y nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1
sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor l
ahf_lm abm 3dnowprefetch cpuid_fault invpcid_single ssbd ibrs ibpb stibp ibrs_enhanc
ed fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx
smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xg
etbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip pku ospke waitpkg av
x512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpi
d cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk arch_lbr amx_bf16 avx5
12_fp16 amx_tile amx_int8 arch_capabilities
Virtualization features:
Hypervisor vendor: KVM
Virtualization type: full
Caches (sum of all):
L1d: 2.1 MiB (45 instances)
L1i: 1.4 MiB (45 instances)
L2: 90 MiB (45 instances)
L3: 97.5 MiB (1 instance)
NUMA:
NUMA node(s): 1
NUMA node0 CPU(s): 0-89
Vulnerabilities:
Gather data sampling: Not affected
Itlb multihit: Not affected
L1tf: Not affected
Mds: Not affected
Meltdown: Not affected
Mmio stale data: Unknown: No mitigations
Retbleed: Not affected
Spec rstack overflow: Not affected
Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Spectre v2: Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Srbds: Not affected
Tsx async abort: Mitigation; TSX disabled
root@iv-ydps7tineowh2yorhj7f:/appdata/sana#

Reproduction

root@ddd064cecf3b:/workspace# curl http://localhost:30000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-M2.1",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": false
  }'
{"id":"d677fc7de40c4ac59e2e513cf69f9381","object":"chat.completion","created":1767416106,"model":"MiniMax-M2.1","choices":[{"index":0,"message":{"role":"assistant","content":"The user says: \" of the response\n\nThey want an answer\n\nHmm, I'm asking\n\n I need to\n\n\n\n\nLet\nThe 用户问\n用户:这是一个\n[\n        {\n            \"user\": \"\\n        }\n    ]\n]\n}\n```\nBut user wants the following:\n```\n\n\nBased\nThe user's request is incomplete. It We�I understand you and. Let me...'s the I'm\n\n\n```python\ndo\n    print(\"Sorry, seems like\")\n```\n\n\n\u0002","reasoning_content":null,"tool_calls":null},"logprobs":null,"finish_reason":"stop","matched_stop":"NaN happened"}],"usage":{"prompt_tokens":40,"total_tokens":144,"completion_tokens":104,"prompt_tokens_details":null,"reasoning_tokens":0},"metadata":{"weight_vers@ddd064cecf3b:/workspace# curl http://localhost:30000/v1/chat/completions \1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-M2.1",
    "messages": [{"role": "user", "content": "你是谁？"}],
    "stream": false
  }'
{"id":"525ff9e6df494d95bbf1669b53a45432","object":"chat.completion","created":1767416154,"model":"MiniMax-M2.1","choices":[{"index":0,"message":{"role":"assistant","content":" ученик\n\nПривет! Ва]\nan prejuí que\n\n\n\n### Крат\nШаги для общения\nПрошу ваы на\n тебя\n\n神经元)\n\n\nThe \n**Шаг 1:** С\n для \n Бет:\n```\n\n框架цн    - Напиши функцию в том,1 Цель\nО, вмес\nТекусту:\n- **тьтерисяз;\nрите        Сейчасерат и другие параметов.\nы ре\n\nЯ\n\n\nХороша, при: 「新历\n\n蛋\n```\nUser\nП\n**\nierungs: Send your.\n```\n\nUser équipement au**ASES\n\n**Те MINI\n опитжа\n    ** кли席**Goal**Крате** Классего и  (\n\n# Те\n**:\n```\n\nЭто**Сначала**\n\n**Эта ира\n```  \n\n```\n\n\n\n** дале谈]\n    - `parse\n\n\n\nUser\n     answer: \nЗкони\n\nОстава\nменююТе. чтобы выражение быout\n```asks for    name: Полказуха\nWe can talk to me\n\n        name**\n

可以看到回答结果和问题毫不相关。
模型下载：https://modelscope.cn/models/MiniMax/MiniMax-M2.1/files
启动命令：kt run m2.1
启动过程没有看到明显错误。

Others

No response

[ErvinXie]

Can you use kt run m2.1 --dry-run? Post the sglang running command here, we have not conduct any test on L20.

[wurenxiaozu]

执行命令：/opt/conda/bin/python -m sglang.launch_server --host 0.0.0.0 --port 30000 --model
/app/ktransformers/MiniMax-M2.1 --kt-weight-path /app/ktransformers/MiniMax-M2.1 --kt-cpuinfer 36
--kt-threadpool-count 1 --kt-num-gpu-experts 58 --kt-method FP8 --kt-gpu-prefill-token-threshold 4096
--attention-backend flashinfer --trust-remote-code --mem-fraction-static 0.8 --chunked-prefill-size 32768
--max-running-requests 16 --max-total-tokens 100000 --watchdog-timeout 3000 --enable-mixed-chunk
--tensor-parallel-size 2 --enable-p2p-check --served-model-name MiniMax-M2.1 --disable-shared-experts-fusion
--fp8-gemm-backend triton --tool-call-parser minimax-m2 --reasoning-parser minimax-append-think

以及：/opt/conda/bin/python -m sglang.launch_server --host 0.0.0.0 --port 30000 --model
/app/ktransformers/MiniMax-M2.1 --kt-weight-path /app/ktransformers/MiniMax-M2.1 --kt-cpuinfer 36
--kt-threadpool-count 1 --kt-num-gpu-experts 58 --kt-method FP8 --kt-gpu-prefill-token-threshold 4096
--attention-backend flashinfer --trust-remote-code --mem-fraction-static 0.8 --chunked-prefill-size 32768
--max-running-requests 16 --max-total-tokens 100000 --watchdog-timeout 3000 --enable-mixed-chunk
--tensor-parallel-size 2 --enable-p2p-check --served-model-name MiniMax-M2.1 --disable-shared-experts-fusion
--fp8-gemm-backend triton --tool-call-parser minimax-m2

结果都一样，都是混乱的返回。