[Question]: 推理时间和报错

[xudong-web]

Question

192.168.254.110 - - [07/Jan/2026 09:57:34] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.05623674392700195
output 69 ←←←← cost: 0.42783212661743164s
output 70 ←←←← cost: 0.5009670257568359s
dual sys step 1.144315242767334
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:36] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.03096461296081543
output 71 ←←←← cost: 0.4664273262023926s
output 72 ←←←← cost: 0.48482704162597656s
dual sys step 1.1698534488677979
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:37] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.037770986557006836
output 73 ←←←← cost: 0.49370431900024414s
output 74 ←←←← cost: 0.45505452156066895s
dual sys step 1.1613214015960693
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:38] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.035863399505615234
output 75 ←←←← cost: 0.7445132732391357s
output 76 ←←←← cost: 0.6807346343994141s
dual sys step 1.7060539722442627
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:40] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.05030107498168945
output 77 ←←←← cost: 0.6840300559997559s
output 78 ←←←← cost: 0.6226468086242676s
dual sys step 1.5666134357452393
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:42] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.09155893325805664
output 79 ←←←← cost: 0.9617910385131836s
output 80 ←←←← cost: 0.7621157169342041s
dual sys step 2.052704095840454
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:47] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.027624130249023438
output 81 ←←←← cost: 0.6731874942779541s
output 82 ←←←← cost: 0.7373943328857422s
dual sys step 1.6695747375488281
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:49] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.028327465057373047
output 83 ←←←← cost: 0.7146737575531006s
output 84 ←←←← cost: 0.8572862148284912s
dual sys step 1.8369107246398926
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:51] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.04564046859741211
output 85 ←←←← cost: 0.9128060340881348s
output 86 ←←←← cost: 0.6549162864685059s
dual sys step 1.8129067420959473
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:53] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.02629566192626953
output 87 ←←←← cost: 0.6465702056884766s
output 88 ←←←← cost: 0.6053526401519775s
dual sys step 1.5387415885925293
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:54] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.03105473518371582
output 89 ←←←← cost: 0.3964419364929199s
output 90 ←←←← cost: 0.3924717903137207s
dual sys step 0.993964672088623
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:57:55] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 4.4431376457214355
read http data cost 0.031417131423950195
output 92 ←←←← cost: 0.737708330154419s
output 93 ←←←← cost: 0.7525179386138916s
[2026-01-07 09:58:01,300] ERROR in app: Exception on /eval_dual [POST]
Traceback (most recent call last):
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/flask/app.py", line 1511, in wsgi_app
response = self.full_dispatch_request()
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/flask/app.py", line 919, in full_dispatch_request
rv = self.handle_user_exception(e)
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/flask/app.py", line 917, in full_dispatch_request
rv = self.dispatch_request()
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/flask/app.py", line 902, in dispatch_request
return self.ensure_sync(self.view_functions[rule.endpoint])(**view_args) # type: ignore[no-any-return]
File "/home/z/InternNav/scripts/realworld/http_internvla_server.py", line 61, in eval_dual
dual_sys_output = agent.step(
File "/home/z/InternNav/internnav/agent/internvla_n1_agent_realworld.py", line 141, in step
self.output_action, self.output_latent, self.output_pixel = self.step_s2(
File "/home/z/InternNav/internnav/agent/internvla_n1_agent_realworld.py", line 231, in step_s2
outputs = self.model.generate(
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/generation/utils.py", line 2460, in generate
result = self._sample(
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/generation/utils.py", line 3419, in _sample
model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
File "/home/z/InternNav/internnav/model/basemodel/internvla_n1/internvla_n1.py", line 118, in prepare_inputs_for_generation
model_inputs = super().prepare_inputs_for_generation(
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py", line 1887, in prepare_inputs_for_generation
model_inputs = super().prepare_inputs_for_generation(
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/generation/utils.py", line 506, in prepare_inputs_for_generation
inputs_embeds, input_ids = self._cache_dependant_input_preparation(
File "/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/generation/utils.py", line 405, in _cache_dependant_input_preparation
or (cache_position[-1] >= input_ids.shape[1]) # Exception 3
IndexError: index -1 is out of bounds for dimension 0 with size 0
192.168.254.110 - - [07/Jan/2026 09:58:01] "POST /eval_dual HTTP/1.1" 500 -
output 94 ←←←← cost: 0.39035511016845703s
dual sys step 1.391981601715088
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:58:01] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.03508400917053223
output 95 ←←←← cost: 0.3931610584259033s
output 96 ←←←← cost: 0.3900918960571289s
dual sys step 1.0014541149139404
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:58:02] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.037839412689208984
output 97 ←←←← cost: 0.3924257755279541s
output 98 ←←←← cost: 0.3930802345275879s
dual sys step 1.0015604496002197
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:58:03] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.03726387023925781
output 99 ←←←← cost: 0.39520955085754395s
output 100 ←←←← cost: 0.403212308883667s
dual sys step 1.017118215560913
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:58:05] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.033812522888183594
output 101 ←←←← cost: 0.4049351215362549s
output 102 ←←←← cost: 0.4033679962158203s
dual sys step 1.0209999084472656
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:58:06] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.03623366355895996
output 103 ←←←← cost: 0.40156078338623047s
output 104 ←←←← cost: 0.39125943183898926s
dual sys step 1.025825023651123
json_output {'discrete_action': [2, 2, 2, 2]}
192.168.254.110 - - [07/Jan/2026 09:58:07] "POST /eval_dual HTTP/1.1" 200 -
read http data cost 0.04863309860229492
output 105 ←←←← cost: 0.3994896411895752s
output 106 ←←←← cost: 0.40450072288513184s
dual sys step 1.0182139873504639
这个推理时间正常吗

[xudong-web]

init reset model!!!
/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/generation/configuration_utils.py:631: UserWarning: do_sample is set to False. However, temperature is set to 0.1 -- this flag is only used in sample-based generation modes. You should set do_sample=True or unset temperature.
warnings.warn(
/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/generation/configuration_utils.py:636: UserWarning: do_sample is set to False. However, top_p is set to 0.001 -- this flag is only used in sample-based generation modes. You should set do_sample=True or unset top_p.
warnings.warn(
/home/z/anaconda3/envs/internnav/lib/python3.9/site-packages/transformers/generation/configuration_utils.py:653: UserWarning: do_sample is set to False. However, top_k is set to 1 -- this flag is only used in sample-based generation modes. You should set do_sample=True or unset top_k.
warnings.warn(
huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
To disable this warning, you can either:
- Avoid using tokenizers before the fork if possible
- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
read http data cost 0.047751665115356445
huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...
To disable this warning, you can either:
- Avoid using tokenizers before the fork if possible
- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)
初始模型加载报错

[xudong-web]

我目前遇到情况是推理速度有时候快（小于1秒），但是有时候会出现一次推理2秒3秒的情况，推理过程还有卡顿情况，以及我刚刚提及的报错，这些导致我的近距离导航可以成功，但是增加左转，右转，就会出现导航失败，遇到目标点椅子，但是好像没有识别到一样，有时候会刻意避开椅子

[yuqiang-yang]

1. 请问是在什么平台进行推理的呢 2. 看起来你的输出几乎都是左转，不太正常 3. 可以检查一下input text给qwen，看看对话对不对，输出纯旋转的时间应该在500ms内

[xudong-web]

1. 请问是在什么平台进行推理的呢 2. 看起来你的输出几乎都是左转，不太正常 3. 可以检查一下input text给qwen，看看对话对不对，输出纯旋转的时间应该在500ms内
   我是在实际机器人上测试，旋转问题不大是因为我把他放在架子上测试导致的，我在测试通信速度，但是我刚刚测试通信http的通信会有很大的卡顿超过3秒，还有报错问题

[xudong-web]

1. 请问是在什么平台进行推理的呢 2. 看起来你的输出几乎都是左转，不太正常 3. 可以检查一下input text给qwen，看看对话对不对，输出纯旋转的时间应该在500ms内

[HTTP] 连接推理服务器失败: HTTPConnectionPool(host='192.168.254.162', port=5802): Read timed out. (read timeout=3.0)
我设置了通信测试，经常出现这种情况

[xudong-web]

我的导航有严重不准确，我的指令是直行到开着的门，然后进去停下来，我请问是什么原因导致呢

[xudong-web]

这个是一部分的导航照片后续导航直接右转直行了

[yuqiang-yang]

通过图片一下子看不太出来为什么。可以继续看一下system2输出的pixel，看看pixel点的是不是对的