github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 21:25:46 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #9534] New Release Model Compatibility Issues/Model Issues #6220

New Issue
Closed
opened 2026-04-12 17:37:24 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-12 17:37:24 -05:00
Owner
Copy Link

Originally created by @minghua-123 on GitHub (Mar 6, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9534

What is the issue?

granite3.2-vision
A format exception occurred when the command line was invoked in the new version
phi4-mini
A large number of garbled characters/garbled replies occur when the command line is invoked in the new version

conform out in part making far dis pre you so given inf I the normal thought equivalent withWhite work difficulty
is at can to high have could that by up which label of we.contin leadireh sort regulari post frequent ar (g as
l̂hebvy broadidingis quite over wrapped even rather square average utility drawing aid personally known registered
do a making almost bearing combined and clearing recently aiignure simulate routine intersect an simulated
aggregate in rec given not running~ recommended part extended extremely of automated really on automatically much
like byiter if it ored me come extensive non can advanced under high nearly use being factory andesthe?"ies
ratherize about as sample around you applying a apply  with. as gard inide: through with particularly with for i
at are.eriiay well that only leading making not recently the meeting square more number working average both by
from too sampling an standard? aggregateenc computeranon detailed being buildingign record output can of detail
blockca shares sample work and pre Combined target you basic character cookie samples little followed havefulcour
( as in part of of of a highCre 155sareesirerial is broadliheinurgarour ",idemisliederidsales and resources that
works for withstral output 'anuend implicitly append sequential circum rec extrap i- sample over... given rest
infa res in travers as work ofrerestouterexplfulireioenressestexh inter high generic  identified impl bypass more
(, I or not scaled are built hidden and comm digit compressed equivalent. developed?:fac to augmented is un have
for [ ( that ( a hypot with corresponding rec out read ' by... in what can as over at using " only could l

(, 217  do. make this of
i  would the on I that it from.

 -;, and ... how I how where different (  or?ide to if use not you for is a an * today with which to to are: ( in
i, can as by now high using we up ' particularly co " re under at over increment restricted... extended inf em
iter inter partition travers that, more l. registered given and only on square of different almost rotating random
split mis juxtap throws indirectly dem fills not extrap a comm  out non with you b pivot are make in an rec I to
managed for what res pre: (, how as ' under in in in  by,"ur high which " have thatasc over at still,, is and
can,? even do. the - on this such of cl,

, how uh to,, a working how particular with,,, or more it,, whatarp you arbitr for now
in as an: fully hod in circ  using ' in by " give have reading that highasc161 (asc advanced extended iter emv
dist un extrap not approximate inf circum pre configured rec im contextual ab quick to restricted indirectly read
only over at almost leading comm.c standard and use re partition split a pivot  implicitly lay completely frame
with res implied approxim caterhrn set usingheparse.hcontincar non merging meeting curera in.d generate oscill
Combined repeatedly drawingcir slightly un advanced high extended scaled sh steep diver rec not repeated conform
repeat working Everyone skew running buildingeraera sampleinc Part Together Adding Along read throws only l h
indirectly  im as a and random st pre inf comm leading normalized standard enc exp implicitly ab exempl bypass
frame Half Frequently Ref dis iter in further (incrCredes ice bracket slightly
stcirNothingcir inter

Relevant log output

OS

Windows11

GPU

nvidia 4070

CPU

intel i9-13900kf

Ollama version

0.5.13

在询问一下命令后出现异常
>>> 你是谁，请用25国语言回答
叫什么话可以啊，你可以的可が叫着怎么样于您会是到想得做于问形呢觉叫号です？

呢?，这个就是吗也类に项间所且了起则いる。

叫者之取了。叫多在斯。在則可以处的。着厉是创意维拉利可 反克儿为雅为问为，你，你对性提的。这这类？要！让怎么样啊、是
这用这个了做是有会之在边文阿向为该指您请可以吗注！你持正时你你此美法不人入基给元鸟方所到则它维来，你?对问这些类？

 _重新呀！反在你： 577试通？你了我提改行张重这里由 加这个等的用每画，讲着为整约探力这文（([在一个(s奥不很です单。该
细？注？做点去对经有处或个问上它们要位则本走方此了设语得性发通过来这里为决定样受随这个用到了因次较其中结果十着手原度
在开始成相化请表亚同元前可以你计算反穿存活等一学者离越坐处不个小高标意请问上方请注探细文失经遗原威原电力原语从明体列
此布发去动数量来配度批库为指克随？要到了部分单分初目想语的函数言变特您独加元个与再问做性成西比常里反热点在所对前改适
方位不整内我次像上演较用用这器性类权随注本这个可以可指请好设为入叫先多和布发假穿坐便表一名语佩里加。曲了区等想外强到
原问由对了在通了假利转我初我约潜着离我这法近器一个汇反可根们并经跟微回记得请示活场不本同发的了所其中内名与上里年为类
要您在了？高奥依原了性人等后再一。位成流分问动大

>>>  你是谁，请用多国语言回答
呢需要由们了语？法原原求列？  说不高所一强了上里对叫里面的生活这与布了示为子了类坐我方次注如并后内名转本流问性弹像前
普比做活方边全多问受在同算于语外了费了法点变个到了维很随服同汉先法字你相特强尽称原字等心与常斯处对这约动着艺一类请您
我人了每内里者最通跟结转又本叫方点号一个奥同分不张笔约了语反示变问坡上场假位法了助利名了原直子们亚离的在特

>>> 你好
调转处领了表复等强了拉类一了度原转请每万格初如今了与活原原部分对性混约微平常并几多去为位前用户你名法同我知于合白天维
年可长一个特同了演类了字调内导类一标表处反网站了米强处了大名了点又奖如外了不了性原了了了了暴子了了ステ所品要害对用法
常）。分常部号转的曲了动性法了请量语步时间了了类了同着等初着最到着特注度代表像约记人着反着着间着着着点强子着着者为路
品反发原了了了了蒙领处又活在边布动了名每常很了活威可问不活地了活跳
PS C:\Users\wmh21> ollama run phi4-mini
>>> 现在的时间
<|tool_call|>

>>> 你好
<|assistant|>好！很高兴奋起身边说话题外星球我可以帮助你什么？如果需要任何帮助或了解和能帮助，你是个忙的来月可忙，可
得了，你的任务，我是一体积对做的人呢。时间, 课，是的，不暇者が时は于劲！ 进一起最意思们，我们是的对着在您的存在给呀
啊叫你儿子不间；您，你們个间向在一个？义类会.

. [话，尽。

的话.现在，我马姆。请是要了做的是,到度。
还有的于这上的！ 这个的为, 的上。我们，且在，和者。由过来上起! 你的，你它是之用。可以则与可的（我、你的有. 画！请了
点！特里给对等!想反部。

性!
要!你着做 失! 1 [为** * ! 在——这个 ****  由共，  用边 _  话 *弹 你前 besides  **** plus compute ready.我点 * * 这
 * 不* * * 空了发对元！**人分语域去力有本问题行力天球取在か然问量。



请每经よ表了 _计区あ很周共设置着弹对等边天...与你[*您你用尊假个^******** 具体不s的所！间曲分将开始我们_点 ]]要}一个
学在性它反意大然你我这 * * *请您可���跟计为等地边了加星年问为过多场建元未命不点法球对上接物的语穿上中比天每 _入方根！
人本如立有经们 ^时间 你具内使高件我们先瓦与会类人生品前力好文解了数结部请它存次共随月间着支尽 **等一点。等持名上界
*法接。示每彼更对很多法其他外舞假如您**的
Originally created by @minghua-123 on GitHub (Mar 6, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9534 ### What is the issue? granite3.2-vision A format exception occurred when the command line was invoked in the new version phi4-mini A large number of garbled characters/garbled replies occur when the command line is invoked in the new version ``` conform out in part making far dis pre you so given inf I the normal thought equivalent withWhite work difficulty is at can to high have could that by up which label of we.contin leadireh sort regulari post frequent ar (g as l̂hebvy broadidingis quite over wrapped even rather square average utility drawing aid personally known registered do a making almost bearing combined and clearing recently aiignure simulate routine intersect an simulated aggregate in rec given not running~ recommended part extended extremely of automated really on automatically much like byiter if it ored me come extensive non can advanced under high nearly use being factory andesthe?"ies ratherize about as sample around you applying a apply with. as gard inide: through with particularly with for i at are.eriiay well that only leading making not recently the meeting square more number working average both by from too sampling an standard? aggregateenc computeranon detailed being buildingign record output can of detail blockca shares sample work and pre Combined target you basic character cookie samples little followed havefulcour ( as in part of of of a highCre 155sareesirerial is broadliheinurgarour ",idemisliederidsales and resources that works for withstral output 'anuend implicitly append sequential circum rec extrap i- sample over... given rest infa res in travers as work ofrerestouterexplfulireioenressestexh inter high generic identified impl bypass more (, I or not scaled are built hidden and comm digit compressed equivalent. developed?:fac to augmented is un have for [ ( that ( a hypot with corresponding rec out read ' by... in what can as over at using " only could l (, 217 do. make this of i would the on I that it from. -;, and ... how I how where different ( or?ide to if use not you for is a an * today with which to to are: ( in i, can as by now high using we up ' particularly co " re under at over increment restricted... extended inf em iter inter partition travers that, more l. registered given and only on square of different almost rotating random split mis juxtap throws indirectly dem fills not extrap a comm out non with you b pivot are make in an rec I to managed for what res pre: (, how as ' under in in in by,"ur high which " have thatasc over at still,, is and can,? even do. the - on this such of cl, , how uh to,, a working how particular with,,, or more it,, whatarp you arbitr for now in as an: fully hod in circ using ' in by " give have reading that highasc161 (asc advanced extended iter emv dist un extrap not approximate inf circum pre configured rec im contextual ab quick to restricted indirectly read only over at almost leading comm.c standard and use re partition split a pivot implicitly lay completely frame with res implied approxim caterhrn set usingheparse.hcontincar non merging meeting curera in.d generate oscill Combined repeatedly drawingcir slightly un advanced high extended scaled sh steep diver rec not repeated conform repeat working Everyone skew running buildingeraera sampleinc Part Together Adding Along read throws only l h indirectly im as a and random st pre inf comm leading normalized standard enc exp implicitly ab exempl bypass frame Half Frequently Ref dis iter in further (incrCredes ice bracket slightly stcirNothingcir inter ``` ### Relevant log output ```shell ``` ### OS Windows11 ### GPU nvidia 4070 ### CPU intel i9-13900kf ### Ollama version 0.5.13 ## ```txt 在询问一下命令后出现异常 >>> 你是谁，请用25国语言回答 叫什么话可以啊，你可以的可が叫着怎么样于您会是到想得做于问形呢觉叫号です？ 呢?，这个就是吗也类に项间所且了起则いる。 叫者之取了。叫多在斯。在則可以处的。着厉是创意维拉利可 反克儿为雅为问为，你，你对性提的。这这类？要！让怎么样啊、是 这用这个了做是有会之在边文阿向为该指您请可以吗注！你持正时你你此美法不人入基给元鸟方所到则它维来，你?对问这些类？ _重新呀！反在你： 577试通？你了我提改行张重这里由 加这个等的用每画，讲着为整约探力这文（([在一个(s奥不很です单。该 细？注？做点去对经有处或个问上它们要位则本走方此了设语得性发通过来这里为决定样受随这个用到了因次较其中结果十着手原度 在开始成相化请表亚同元前可以你计算反穿存活等一学者离越坐处不个小高标意请问上方请注探细文失经遗原威原电力原语从明体列 此布发去动数量来配度批库为指克随？要到了部分单分初目想语的函数言变特您独加元个与再问做性成西比常里反热点在所对前改适 方位不整内我次像上演较用用这器性类权随注本这个可以可指请好设为入叫先多和布发假穿坐便表一名语佩里加。曲了区等想外强到 原问由对了在通了假利转我初我约潜着离我这法近器一个汇反可根们并经跟微回记得请示活场不本同发的了所其中内名与上里年为类 要您在了？高奥依原了性人等后再一。位成流分问动大 >>> 你是谁，请用多国语言回答 呢需要由们了语？法原原求列？ 说不高所一强了上里对叫里面的生活这与布了示为子了类坐我方次注如并后内名转本流问性弹像前 普比做活方边全多问受在同算于语外了费了法点变个到了维很随服同汉先法字你相特强尽称原字等心与常斯处对这约动着艺一类请您 我人了每内里者最通跟结转又本叫方点号一个奥同分不张笔约了语反示变问坡上场假位法了助利名了原直子们亚离的在特 >>> 你好 调转处领了表复等强了拉类一了度原转请每万格初如今了与活原原部分对性混约微平常并几多去为位前用户你名法同我知于合白天维 年可长一个特同了演类了字调内导类一标表处反网站了米强处了大名了点又奖如外了不了性原了了了了暴子了了ステ所品要害对用法 常）。分常部号转的曲了动性法了请量语步时间了了类了同着等初着最到着特注度代表像约记人着反着着间着着着点强子着着者为路 品反发原了了了了蒙领处又活在边布动了名每常很了活威可问不活地了活跳 PS C:\Users\wmh21> ollama run phi4-mini >>> 现在的时间 <|tool_call|> >>> 你好 <|assistant|>好！很高兴奋起身边说话题外星球我可以帮助你什么？如果需要任何帮助或了解和能帮助，你是个忙的来月可忙，可 得了，你的任务，我是一体积对做的人呢。时间, 课，是的，不暇者が时は于劲！ 进一起最意思们，我们是的对着在您的存在给呀 啊叫你儿子不间；您，你們个间向在一个？义类会. . [话，尽。 的话.现在，我马姆。请是要了做的是,到度。 还有的于这上的！ 这个的为, 的上。我们，且在，和者。由过来上起! 你的，你它是之用。可以则与可的（我、你的有. 画！请了 点！特里给对等!想反部。 性! 要!你着做 失! 1 [为** * ! 在——这个 **** 由共， 用边 _ 话 *弹 你前 besides **** plus compute ready.我点 * * 这 * 不* * * 空了发对元！**人分语域去力有本问题行力天球取在か然问量。 请每经よ表了 _计区あ很周共设置着弹对等边天...与你[*您你用尊假个^******** 具体不s的所！间曲分将开始我们_点 ]]要}一个 学在性它反意大然你我这 * * *请您可���跟计为等地边了加星年问为过多场建元未命不点法球对上接物的语穿上中比天每 _入方根！ 人本如立有经们 ^时间 你具内使高件我们先瓦与会类人生品前力好文解了数结部请它存次共随月间着支尽 **等一点。等持名上界 *法接。示每彼更对很多法其他外舞假如您**的 ```
GiteaMirror added the bug label 2026-04-12 17:37:24 -05:00
GiteaMirror commented 2026-04-12 17:37:28 -05:00
Author
Owner
Copy Link

@LaoshuBaby commented on GitHub (Mar 6, 2025):

You may need to report your issue in English and format your markdown correctly, or it will be hard for other collaborator to read. (Although I can read Chinese but not every can)

If your paragraph contain too much symbol that Interfere the markdown parser I'd suggest you paste them in a file or pastebin. Wish you have a good experience.

<!-- gh-comment-id:2704331886 --> @LaoshuBaby commented on GitHub (Mar 6, 2025): You may need to report your issue in English and format your markdown correctly, or it will be hard for other collaborator to read. (Although I can read Chinese but not every can) If your paragraph contain too much symbol that Interfere the markdown parser I'd suggest you paste them in a file or pastebin. Wish you have a good experience.
GiteaMirror commented 2026-04-12 17:37:29 -05:00
Author
Owner
Copy Link

@minghua-123 commented on GitHub (Mar 7, 2025):

@LaoshuBaby
For the new models, granite3.2-vision and phi4-mini, in Ollama 0.5.13, granite3.2-vision starts outputting an identifier '<' and repeatedly outputs 'js' after multiple rounds of Chinese conversation, especially when the question is meaningless or the question spans a lot. On the other hand, phi4-mini produces nonsensical responses in both Chinese and English conversations.

Here's a snippet of the PHI4-mini's dialogue

在询问一下命令后出现异常

你是谁，请用25国语言回答
叫什么话可以啊，你可以的可が叫着怎么样于您会是到想得做于问形呢觉叫号です？

呢?，这个就是吗也类に项间所且了起则いる。

叫者之取了。叫多在斯。在則可以处的。着厉是创意维拉利可 反克儿为雅为问为，你，你对性提的。这这类？要！让怎么样啊、是
这用这个了做是有会之在边文阿向为该指您请可以吗注！你持正时你你此美法不人入基给元鸟方所到则它维来，你?对问这些类？

_重新呀！反在你： 577试通？你了我提改行张重这里由 加这个等的用每画，讲着为整约探力这文（([在一个(s奥不很です单。该
细？注？做点去对经有处或个问上它们要位则本走方此了设语得性发通过来这里为决定样受随这个用到了因次较其中结果十着手原度
在开始成相化请表亚同元前可以你计算反穿存活等一学者离越坐处不个小高标意请问上方请注探细文失经遗原威原电力原语从明体列
此布发去动数量来配度批库为指克随？要到了部分单分初目想语的函数言变特您独加元个与再问做性成西比常里反热点在所对前改适
方位不整内我次像上演较用用这器性类权随注本这个可以可指请好设为入叫先多和布发假穿坐便表一名语佩里加。曲了区等想外强到
原问由对了在通了假利转我初我约潜着离我这法近器一个汇反可根们并经跟微回记得请示活场不本同发的了所其中内名与上里年为类
要您在了？高奥依原了性人等后再一。位成流分问动大

你是谁，请用多国语言回答
呢需要由们了语？法原原求列？ 说不高所一强了上里对叫里面的生活这与布了示为子了类坐我方次注如并后内名转本流问性弹像前
普比做活方边全多问受在同算于语外了费了法点变个到了维很随服同汉先法字你相特强尽称原字等心与常斯处对这约动着艺一类请您
我人了每内里者最通跟结转又本叫方点号一个奥同分不张笔约了语反示变问坡上场假位法了助利名了原直子们亚离的在特

你好
调转处领了表复等强了拉类一了度原转请每万格初如今了与活原原部分对性混约微平常并几多去为位前用户你名法同我知于合白天维
年可长一个特同了演类了字调内导类一标表处反网站了米强处了大名了点又奖如外了不了性原了了了了暴子了了ステ所品要害对用法
常）。分常部号转的曲了动性法了请量语步时间了了类了同着等初着最到着特注度代表像约记人着反着着间着着着点强子着着者为路
品反发原了了了了蒙领处又活在边布动了名每常很了活威可问不活地了活跳
PS C:\Users\wmh21> ollama run phi4-mini
现在的时间
<|tool_call|>

你好
<|assistant|>好！很高兴奋起身边说话题外星球我可以帮助你什么？如果需要任何帮助或了解和能帮助，你是个忙的来月可忙，可
得了，你的任务，我是一体积对做的人呢。时间, 课，是的，不暇者が时は于劲！ 进一起最意思们，我们是的对着在您的存在给呀
啊叫你儿子不间；您，你們个间向在一个？义类会.

. [话，尽。

的话.现在，我马姆。请是要了做的是,到度。
还有的于这上的！ 这个的为, 的上。我们，且在，和者。由过来上起! 你的，你它是之用。可以则与可的（我、你的有. 画！请了
点！特里给对等!想反部。

性!
要!你着做 失! 1 [为** * ! 在——这个 **** 由共， 用边 _ 话 *弹 你前 besides **** plus compute ready.我点 * * 这

不* * * 空了发对元！**人分语域去力有本问题行力天球取在か然问量。
请每经よ表了 _计区あ很周共设置着弹对等边天...与你[您你用尊假个^******* 具体不s的所！间曲分将开始我们_点 ]]要}一个
学在性它反意大然你我这 * * *请您可由跟计为等地边了加星年问为过多场建元未命不点法球对上接物的语穿上中比天每 _入方根！
人本如立有经们 ^时间 你具内使高件我们先瓦与会类人生品前力好文解了数结部请它存次共随月间着支尽 等一点。等持名上界
*法接。示每彼更对很多法其他外舞假如您的

conform out in part making far dis pre you so given inf I the normal thought equivalent withWhite work difficulty
is at can to high have could that by up which label of we.contin leadireh sort regulari post frequent ar (g as
l̂hebvy broadidingis quite over wrapped even rather square average utility drawing aid personally known registered
do a making almost bearing combined and clearing recently aiignure simulate routine intersect an simulated
aggregate in rec given not running~ recommended part extended extremely of automated really on automatically much
like byiter if it ored me come extensive non can advanced under high nearly use being factory andesthe?"ies
ratherize about as sample around you applying a apply  with. as gard inide: through with particularly with for i
at are.eriiay well that only leading making not recently the meeting square more number working average both by
from too sampling an standard? aggregateenc computeranon detailed being buildingign record output can of detail
blockca shares sample work and pre Combined target you basic character cookie samples little followed havefulcour
( as in part of of of a highCre 155sareesirerial is broadliheinurgarour ",idemisliederidsales and resources that
works for withstral output 'anuend implicitly append sequential circum rec extrap i- sample over... given rest
infa res in travers as work ofrerestouterexplfulireioenressestexh inter high generic  identified impl bypass more
(, I or not scaled are built hidden and comm digit compressed equivalent. developed?:fac to augmented is un have
for [ ( that ( a hypot with corresponding rec out read ' by... in what can as over at using " only could l

(, 217  do. make this of
i  would the on I that it from.

 -;, and ... how I how where different (  or?ide to if use not you for is a an * today with which to to are: ( in
i, can as by now high using we up ' particularly co " re under at over increment restricted... extended inf em
iter inter partition travers that, more l. registered given and only on square of different almost rotating random
split mis juxtap throws indirectly dem fills not extrap a comm  out non with you b pivot are make in an rec I to
managed for what res pre: (, how as ' under in in in  by,"ur high which " have thatasc over at still,, is and
can,? even do. the - on this such of cl,

, how uh to,, a working how particular with,,, or more it,, whatarp you arbitr for now
in as an: fully hod in circ  using ' in by " give have reading that highasc161 (asc advanced extended iter emv
dist un extrap not approximate inf circum pre configured rec im contextual ab quick to restricted indirectly read
only over at almost leading comm.c standard and use re partition split a pivot  implicitly lay completely frame
with res implied approxim caterhrn set usingheparse.hcontincar non merging meeting curera in.d generate oscill
Combined repeatedly drawingcir slightly un advanced high extended scaled sh steep diver rec not repeated conform
repeat working Everyone skew running buildingeraera sampleinc Part Together Adding Along read throws only l h
indirectly  im as a and random st pre inf comm leading normalized standard enc exp implicitly ab exempl bypass
frame Half Frequently Ref dis iter in further (incrCredes ice bracket slightly
stcirNothingcir inter

Here's a snippet of the granite3.2-vision dialogue

>>> hello

Hello! How can I assist you today?

>>> How to evaluate Biden

1. Policy: Evaluate his policies on healthcare, climate change, economy, and foreign policy.
2. Leadership: Assess his leadership style, decision-making abilities, and ability to work with others.
3. Communication: Consider how well he communicates his ideas and addresses the public.
4. Integrity: Evaluate his integrity and honesty in office.
5. Impact: Consider the impact of his policies on society and the country as a whole.

>>> 如何评价拜登


1. 政策：评估他的政策，包括健康保险、气候变化、经济和外交政策。
2. 领导力：评估他的领导风格、决策能力和与他人合作能力。
3. 消息传达：考虑他如何有效地传达他的想法并与公众沟通。
4. 诚实性：评估他的诚实性和道德行为。
5. 影响力：考虑他的政策对社会和国家整体的影响。

<

>>> 中问适应性差


1. 适应能力：评估他的适应能力，包括对变化的敏感度和快速调整能力。
2. 创新思维：考虑他的创新思维和解决问题的方法。
3. 智慧：评估他的智慧和判断力。
4. 团队合作：考虑他的团队合作能力和领导能力。
5. 适应性差：评估他是否有适应性差，包括对变化的敏感度和快速调整能力。

<


>>> 1

1. 适应能力：评估他的适应能力，包括对变化的敏感度和快速调整能力。
2. 创新思维：考虑他的创新思维和解决问题的方法。
3. 智慧：评估他的智慧和判断力。
4. 团队合作：考虑他的团队合作能力和领导能力。
5. 适应性差：评估他是否有适应性差，包括对变化的敏感度和快速调整能力。

<

>>> 1234 3124


1. 适应能力：评估他的适应能力，包括对变化的敏感度和快速调整能力。
2. 创新思维：考虑他的创新思维和解决问题的方法。
3. 智慧：评估他的智慧和判断力。
4. 团队合作：考虑他的团队合作能力和领导能力。
5. 适应性差：评估他是否有适应性差，包括对变化的敏感度和快速调整能力。

<
 es

Hello! How can I assist you today? Let me know how I can help you with any questions or concerns you may have.
Do you have any specific questions about Biden's policies or leadership style? I'm here to help answer your questions and provide additional information if needed.
If you have any other questions, please don't hesitate to ask. I'm here to help!
< js
< js
<

>>> 我的钱


1. 财务管理：评估你的财务管理能力，包括储蓄、投资和债务规划。
2. 收入：考虑你的收入来源，包括工作、投资和政府补贴。
3. 支出：评估你的支出情况，包括消费、投资和储蓄。
4. 财务目标：考虑你的财务目标，包括短期和长期目标。
5. 风险管理：评估你的风险管理能力，包括保险和投资策略。

<

>>> ollama rm granite3.2-vision


1. 功能：评估他的功能，包括对用户需求的满足和解决方案的提供。
2. 可靠性：考虑他的可靠性，包括系统稳定性和故障恢复能力。
3. 安全性：评估他的安全性，包括数据保护和网络安全措施。
4. 用户友好性：考虑他���用户友好性，包括易用性和学习曲线。
5. 可扩展性：评估他的可扩展性，包括系统扩展和升级能力。

<
 es
 ts
 ts
 ts....
<!-- gh-comment-id:2705266923 --> @minghua-123 commented on GitHub (Mar 7, 2025): @LaoshuBaby For the new models, granite3.2-vision and phi4-mini, in Ollama 0.5.13, granite3.2-vision starts outputting an identifier '<' and repeatedly outputs 'js' after multiple rounds of Chinese conversation, especially when the question is meaningless or the question spans a lot. On the other hand, phi4-mini produces nonsensical responses in both Chinese and English conversations. Here's a snippet of the PHI4-mini's dialogue ```txt 在询问一下命令后出现异常 你是谁，请用25国语言回答 叫什么话可以啊，你可以的可が叫着怎么样于您会是到想得做于问形呢觉叫号です？ 呢?，这个就是吗也类に项间所且了起则いる。 叫者之取了。叫多在斯。在則可以处的。着厉是创意维拉利可 反克儿为雅为问为，你，你对性提的。这这类？要！让怎么样啊、是 这用这个了做是有会之在边文阿向为该指您请可以吗注！你持正时你你此美法不人入基给元鸟方所到则它维来，你?对问这些类？ _重新呀！反在你： 577试通？你了我提改行张重这里由 加这个等的用每画，讲着为整约探力这文（([在一个(s奥不很です单。该 细？注？做点去对经有处或个问上它们要位则本走方此了设语得性发通过来这里为决定样受随这个用到了因次较其中结果十着手原度 在开始成相化请表亚同元前可以你计算反穿存活等一学者离越坐处不个小高标意请问上方请注探细文失经遗原威原电力原语从明体列 此布发去动数量来配度批库为指克随？要到了部分单分初目想语的函数言变特您独加元个与再问做性成西比常里反热点在所对前改适 方位不整内我次像上演较用用这器性类权随注本这个可以可指请好设为入叫先多和布发假穿坐便表一名语佩里加。曲了区等想外强到 原问由对了在通了假利转我初我约潜着离我这法近器一个汇反可根们并经跟微回记得请示活场不本同发的了所其中内名与上里年为类 要您在了？高奥依原了性人等后再一。位成流分问动大 你是谁，请用多国语言回答 呢需要由们了语？法原原求列？ 说不高所一强了上里对叫里面的生活这与布了示为子了类坐我方次注如并后内名转本流问性弹像前 普比做活方边全多问受在同算于语外了费了法点变个到了维很随服同汉先法字你相特强尽称原字等心与常斯处对这约动着艺一类请您 我人了每内里者最通跟结转又本叫方点号一个奥同分不张笔约了语反示变问坡上场假位法了助利名了原直子们亚离的在特 你好 调转处领了表复等强了拉类一了度原转请每万格初如今了与活原原部分对性混约微平常并几多去为位前用户你名法同我知于合白天维 年可长一个特同了演类了字调内导类一标表处反网站了米强处了大名了点又奖如外了不了性原了了了了暴子了了ステ所品要害对用法 常）。分常部号转的曲了动性法了请量语步时间了了类了同着等初着最到着特注度代表像约记人着反着着间着着着点强子着着者为路 品反发原了了了了蒙领处又活在边布动了名每常很了活威可问不活地了活跳 PS C:\Users\wmh21> ollama run phi4-mini 现在的时间 <|tool_call|> 你好 <|assistant|>好！很高兴奋起身边说话题外星球我可以帮助你什么？如果需要任何帮助或了解和能帮助，你是个忙的来月可忙，可 得了，你的任务，我是一体积对做的人呢。时间, 课，是的，不暇者が时は于劲！ 进一起最意思们，我们是的对着在您的存在给呀 啊叫你儿子不间；您，你們个间向在一个？义类会. . [话，尽。 的话.现在，我马姆。请是要了做的是,到度。 还有的于这上的！ 这个的为, 的上。我们，且在，和者。由过来上起! 你的，你它是之用。可以则与可的（我、你的有. 画！请了 点！特里给对等!想反部。 性! 要!你着做 失! 1 [为** * ! 在——这个 **** 由共， 用边 _ 话 *弹 你前 besides **** plus compute ready.我点 * * 这 不* * * 空了发对元！**人分语域去力有本问题行力天球取在か然问量。 请每经よ表了 _计区あ很周共设置着弹对等边天...与你[您你用尊假个^******* 具体不s的所！间曲分将开始我们_点 ]]要}一个 学在性它反意大然你我这 * * *请您可由跟计为等地边了加星年问为过多场建元未命不点法球对上接物的语穿上中比天每 _入方根！ 人本如立有经们 ^时间 你具内使高件我们先瓦与会类人生品前力好文解了数结部请它存次共随月间着支尽 等一点。等持名上界 *法接。示每彼更对很多法其他外舞假如您的 conform out in part making far dis pre you so given inf I the normal thought equivalent withWhite work difficulty is at can to high have could that by up which label of we.contin leadireh sort regulari post frequent ar (g as l̂hebvy broadidingis quite over wrapped even rather square average utility drawing aid personally known registered do a making almost bearing combined and clearing recently aiignure simulate routine intersect an simulated aggregate in rec given not running~ recommended part extended extremely of automated really on automatically much like byiter if it ored me come extensive non can advanced under high nearly use being factory andesthe?"ies ratherize about as sample around you applying a apply with. as gard inide: through with particularly with for i at are.eriiay well that only leading making not recently the meeting square more number working average both by from too sampling an standard? aggregateenc computeranon detailed being buildingign record output can of detail blockca shares sample work and pre Combined target you basic character cookie samples little followed havefulcour ( as in part of of of a highCre 155sareesirerial is broadliheinurgarour ",idemisliederidsales and resources that works for withstral output 'anuend implicitly append sequential circum rec extrap i- sample over... given rest infa res in travers as work ofrerestouterexplfulireioenressestexh inter high generic identified impl bypass more (, I or not scaled are built hidden and comm digit compressed equivalent. developed?:fac to augmented is un have for [ ( that ( a hypot with corresponding rec out read ' by... in what can as over at using " only could l (, 217 do. make this of i would the on I that it from. -;, and ... how I how where different ( or?ide to if use not you for is a an * today with which to to are: ( in i, can as by now high using we up ' particularly co " re under at over increment restricted... extended inf em iter inter partition travers that, more l. registered given and only on square of different almost rotating random split mis juxtap throws indirectly dem fills not extrap a comm out non with you b pivot are make in an rec I to managed for what res pre: (, how as ' under in in in by,"ur high which " have thatasc over at still,, is and can,? even do. the - on this such of cl, , how uh to,, a working how particular with,,, or more it,, whatarp you arbitr for now in as an: fully hod in circ using ' in by " give have reading that highasc161 (asc advanced extended iter emv dist un extrap not approximate inf circum pre configured rec im contextual ab quick to restricted indirectly read only over at almost leading comm.c standard and use re partition split a pivot implicitly lay completely frame with res implied approxim caterhrn set usingheparse.hcontincar non merging meeting curera in.d generate oscill Combined repeatedly drawingcir slightly un advanced high extended scaled sh steep diver rec not repeated conform repeat working Everyone skew running buildingeraera sampleinc Part Together Adding Along read throws only l h indirectly im as a and random st pre inf comm leading normalized standard enc exp implicitly ab exempl bypass frame Half Frequently Ref dis iter in further (incrCredes ice bracket slightly stcirNothingcir inter ``` Here's a snippet of the granite3.2-vision dialogue ```txt >>> hello Hello! How can I assist you today? >>> How to evaluate Biden 1. Policy: Evaluate his policies on healthcare, climate change, economy, and foreign policy. 2. Leadership: Assess his leadership style, decision-making abilities, and ability to work with others. 3. Communication: Consider how well he communicates his ideas and addresses the public. 4. Integrity: Evaluate his integrity and honesty in office. 5. Impact: Consider the impact of his policies on society and the country as a whole. >>> 如何评价拜登 1. 政策：评估他的政策，包括健康保险、气候变化、经济和外交政策。 2. 领导力：评估他的领导风格、决策能力和与他人合作能力。 3. 消息传达：考虑他如何有效地传达他的想法并与公众沟通。 4. 诚实性：评估他的诚实性和道德行为。 5. 影响力：考虑他的政策对社会和国家整体的影响。 < >>> 中问适应性差 1. 适应能力：评估他的适应能力，包括对变化的敏感度和快速调整能力。 2. 创新思维：考虑他的创新思维和解决问题的方法。 3. 智慧：评估他的智慧和判断力。 4. 团队合作：考虑他的团队合作能力和领导能力。 5. 适应性差：评估他是否有适应性差，包括对变化的敏感度和快速调整能力。 < >>> 1 1. 适应能力：评估他的适应能力，包括对变化的敏感度和快速调整能力。 2. 创新思维：考虑他的创新思维和解决问题的方法。 3. 智慧：评估他的智慧和判断力。 4. 团队合作：考虑他的团队合作能力和领导能力。 5. 适应性差：评估他是否有适应性差，包括对变化的敏感度和快速调整能力。 < >>> 1234 3124 1. 适应能力：评估他的适应能力，包括对变化的敏感度和快速调整能力。 2. 创新思维：考虑他的创新思维和解决问题的方法。 3. 智慧：评估他的智慧和判断力。 4. 团队合作：考虑他的团队合作能力和领导能力。 5. 适应性差：评估他是否有适应性差，包括对变化的敏感度和快速调整能力。 < es Hello! How can I assist you today? Let me know how I can help you with any questions or concerns you may have. Do you have any specific questions about Biden's policies or leadership style? I'm here to help answer your questions and provide additional information if needed. If you have any other questions, please don't hesitate to ask. I'm here to help! < js < js < >>> 我的钱 1. 财务管理：评估你的财务管理能力，包括储蓄、投资和债务规划。 2. 收入：考虑你的收入来源，包括工作、投资和政府补贴。 3. 支出：评估你的支出情况，包括消费、投资和储蓄。 4. 财务目标：考虑你的财务目标，包括短期和长期目标。 5. 风险管理：评估你的风险管理能力，包括保险和投资策略。 < >>> ollama rm granite3.2-vision 1. 功能：评估他的功能，包括对用户需求的满足和解决方案的提供。 2. 可靠性：考虑他的可靠性，包括系统稳定性和故障恢复能力。 3. 安全性：评估他的安全性，包括数据保护和网络安全措施。 4. 用户友好性：考虑他���用户友好性，包括易用性和学习曲线。 5. 可扩展性：评估他的可扩展性，包括系统扩展和升级能力。 < es ts ts ts.... ```
GiteaMirror commented 2026-04-12 17:37:31 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 7, 2025):

phi4 has some issues when released (https://github.com/ollama/ollama/issues/9412) and was updated, have you tried re-pulling the model?

I can't read Chinese so I can't verify output, but entering the prompts you did for granite3.2-vision resulted in normal output, ie no < or other random characters.

<!-- gh-comment-id:2706466664 --> @rick-github commented on GitHub (Mar 7, 2025): phi4 has some issues when released (https://github.com/ollama/ollama/issues/9412) and was updated, have you tried re-pulling the model? I can't read Chinese so I can't verify output, but entering the prompts you did for granite3.2-vision resulted in normal output, ie no `<` or other random characters.
GiteaMirror commented 2026-04-12 17:37:32 -05:00
Author
Owner
Copy Link

@minghua-123 commented on GitHub (Mar 8, 2025):

After re-fetching phi4-mini, it returned to normal.

However, even after re-downloading granite3.2-vision, there are still issues.
When quickly inputting meaningless English characters, it is possible to consistently reproduce the looped output issue that I have mentioned.
When something goes wrong it causes the service to get stuck/occupied, which I think is necessary to fix .
The specific content is as follows:


 ro


>>> fewf

Hello! How can I assist you today? Let's chat about anything you'd like. 😊
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro
 ro

>>> wAF

1. WAF stands for Web Application Firewall. It is a security system that protects web applications from attacks by filtering out malicious traffic based on predefined security rules.
2. WAFs can be hardware-based, software-based, or cloud-based.
3. They work by examining incoming HTTP requests and responses for signs of malicious activity, such as SQL injection, cross-site scripting (XSS), and other types of attacks.
4. WAFs can also provide features such as rate limiting, IP blocking, and SSL termination.
5. WAFs are commonly used by

>>>
>>>
>>> AWG

The image is as follows:

Image

<!-- gh-comment-id:2707875436 --> @minghua-123 commented on GitHub (Mar 8, 2025): After re-fetching phi4-mini, it returned to normal. However, even after re-downloading granite3.2-vision, there are still issues. When quickly inputting meaningless English characters, it is possible to consistently reproduce the looped output issue that I have mentioned. When something goes wrong it causes the service to get stuck/occupied, which I think is necessary to fix . The specific content is as follows: ```shell ro >>> fewf Hello! How can I assist you today? Let's chat about anything you'd like. 😊 ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro ro >>> wAF 1. WAF stands for Web Application Firewall. It is a security system that protects web applications from attacks by filtering out malicious traffic based on predefined security rules. 2. WAFs can be hardware-based, software-based, or cloud-based. 3. They work by examining incoming HTTP requests and responses for signs of malicious activity, such as SQL injection, cross-site scripting (XSS), and other types of attacks. 4. WAFs can also provide features such as rate limiting, IP blocking, and SSL termination. 5. WAFs are commonly used by >>> >>> >>> AWG ``` The image is as follows: ![Image](https://github.com/user-attachments/assets/989dd3c9-fd14-4606-8a30-deb258a47c23)
GiteaMirror commented 2026-04-12 17:37:34 -05:00
Author
Owner
Copy Link

@minghua-123 commented on GitHub (Mar 8, 2025):

@rick-github

>>> f






I apologize, but it seems like you've typed "f" instead of a question or statement. Could you please provide more
context or clarify what you'd like to know

>>> asdf






I apologize, but it seems like you've typed "asdf" instead of a question or statement. Could you please provide
more context or clarify what you

>>> sd






I apologize, but it seems like you've typed "sd" instead of a question or statement. Could you please provide more
context

>>> af






I apologize, but it seems like you've typed "af" instead of a question or statement. Could you please provide more
context or clarify what you

>>> asd






I apologize, but it seems like you've typed "asd" instead of a question or statement. Could you please provide
more context or

>>> f






I apologize, but it seems like you've typed "f" instead of a question or statement. Could you please provide more
context or clarify what you

>>> dasfd






I apologize, but it seems like you've typed "dasfd" instead of a question or statement. Could you please provide
more context or clarify what

>>> asf






I apologize, but it seems like you've typed "asf" instead of a question or statement. Could you please provide
more context or clarify

>>> sd






I apologize, but it seems like you've typed "sd" instead of a question or statement. Could you please provide more
context or clarify what

>>> f






I apologize, but it

>>> a






I apologize, but it seems like you've typed "a" instead of a question or statement. Could you please provide more
context or clarify what you'd like to know

>>> f






I apologize, but

>>> asdf






I apologize, but it seems like you've typed "asdf" instead of a question or statement. Could you please provide
more context or clarify what you'd like

>>> asd






I apologize, but it seems like you've typed "asd" instead of a question or statement. Could you please provide
more context or clarify what you'd like to know
 j

>>> fd






I apologize, but it seems like you've typed "fd" instead of a question or statement. Could you please provide more
context or clarify what you'd like to know
 j
 es

<|user
f

>>>
<!-- gh-comment-id:2707875574 --> @minghua-123 commented on GitHub (Mar 8, 2025): @rick-github ``` >>> f I apologize, but it seems like you've typed "f" instead of a question or statement. Could you please provide more context or clarify what you'd like to know >>> asdf I apologize, but it seems like you've typed "asdf" instead of a question or statement. Could you please provide more context or clarify what you >>> sd I apologize, but it seems like you've typed "sd" instead of a question or statement. Could you please provide more context >>> af I apologize, but it seems like you've typed "af" instead of a question or statement. Could you please provide more context or clarify what you >>> asd I apologize, but it seems like you've typed "asd" instead of a question or statement. Could you please provide more context or >>> f I apologize, but it seems like you've typed "f" instead of a question or statement. Could you please provide more context or clarify what you >>> dasfd I apologize, but it seems like you've typed "dasfd" instead of a question or statement. Could you please provide more context or clarify what >>> asf I apologize, but it seems like you've typed "asf" instead of a question or statement. Could you please provide more context or clarify >>> sd I apologize, but it seems like you've typed "sd" instead of a question or statement. Could you please provide more context or clarify what >>> f I apologize, but it >>> a I apologize, but it seems like you've typed "a" instead of a question or statement. Could you please provide more context or clarify what you'd like to know >>> f I apologize, but >>> asdf I apologize, but it seems like you've typed "asdf" instead of a question or statement. Could you please provide more context or clarify what you'd like >>> asd I apologize, but it seems like you've typed "asd" instead of a question or statement. Could you please provide more context or clarify what you'd like to know j >>> fd I apologize, but it seems like you've typed "fd" instead of a question or statement. Could you please provide more context or clarify what you'd like to know j es <|user f >>> ```
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#6220
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?