rpc error: code = Unknown desc = unimplemented

[allenhaozi]

What went wrong? Settings?

quay.io/go-skynet/local-ai:master-cublas-cuda11

request:

{
    "model": "llama-7b-hf",
    "messages": [
        {
            "role": "user",
            "content": "Hello! What is your name?"
        }
    ]
}

response:

{
    "error": {
        "code": 500,
        "message": "rpc error: code = Unknown desc = unimplemented",
        "type": ""
    }
}

log:

@@@@@
Skipping rebuild
@@@@@
If you are experiencing issues with the pre-compiled builds, try setting REBUILD=true
If you are still experiencing issues with the build, try setting CMAKE_ARGS and disable the instructions set as needed:
CMAKE_ARGS="-DLLAMA_F16C=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF"
see the documentation at: https://localai.io/basics/build/index.html
Note: See also https://github.com/go-skynet/LocalAI/issues/288
@@@@@
CPU info:
model name      : Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb cat_l3 cdp_l3 invpcid_single intel_ppin intel_pt tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm rdt_a rdseed adx smap xsaveopt cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts
CPU:    AVX    found OK
CPU:    AVX2   found OK
CPU: no AVX512 found
@@@@@
ESC[90m5:38AMESC[0m ESC[33mDBGESC[0m no galleries to load
ESC[90m5:38AMESC[0m ESC[32mINFESC[0m Starting LocalAI using 4 threads, with models path: /llm-model-volume
ESC[90m5:38AMESC[0m ESC[32mINFESC[0m LocalAI version: 12fe093 (12fe0932c41246914e455c4175269a431fb8cf60)
ESC[90m5:38AMESC[0m ESC[33mDBGESC[0m Extracting backend assets files to /tmp/localai/backend_data

 ┌───────────────────────────────────────────────────┐ 
 │                   Fiber v2.48.0                   │ 
 │               http://127.0.0.1:8080               │ 
 │       (bound on host 0.0.0.0 and port 8080)       │ 
 │                                                   │ 
 │ Handlers ............ 32  Processes ........... 1 │ 
 │ Prefork ....... Disabled  PID ................ 14 │ 
 └───────────────────────────────────────────────────┘ 

6:21AM DBG Request received: 
6:21AM DBG Configuration read: &{PredictionOptions:{Model:llama-7b-hf Language: N:0 TopP:0.7 TopK:80 Temperature:0.9 Maxtokens:512 Echo:false Batch:0 F16:false IgnoreEOS:false RepeatPenalty:0 Keep:0 MirostatETA:0 MirostatTAU:0 Mirostat:0 FrequencyPenalty:0 TFZ:0 TypicalP:0 Seed:0} Name: StopWords:[] Cutstrings:[] TrimSpace:[] ContextSize:512 F16:false NUMA:false Threads:4 Debug:true Roles:map[] Embeddings:false Backend: TemplateConfig:{Chat: ChatMessage: Completion: Edit: Functions:} MirostatETA:0 MirostatTAU:0 Mirostat:0 NGPULayers:0 MMap:false MMlock:false LowVRAM:false TensorSplit: MainGPU: ImageGenerationAssets: PromptCachePath: PromptCacheAll:false PromptCacheRO:false Grammar: PromptStrings:[] InputStrings:[] InputToken:[] functionCallString: functionCallNameString: FunctionsConfig:{DisableNoAction:false NoActionFunctionName: NoActionDescriptionName:} SystemPrompt:}
6:21AM DBG Parameters: &{PredictionOptions:{Model:llama-7b-hf Language: N:0 TopP:0.7 TopK:80 Temperature:0.9 Maxtokens:512 Echo:false Batch:0 F16:false IgnoreEOS:false RepeatPenalty:0 Keep:0 MirostatETA:0 MirostatTAU:0 Mirostat:0 FrequencyPenalty:0 TFZ:0 TypicalP:0 Seed:0} Name: StopWords:[] Cutstrings:[] TrimSpace:[] ContextSize:512 F16:false NUMA:false Threads:4 Debug:true Roles:map[] Embeddings:false Backend: TemplateConfig:{Chat: ChatMessage: Completion: Edit: Functions:} MirostatETA:0 MirostatTAU:0 Mirostat:0 NGPULayers:0 MMap:false MMlock:false LowVRAM:false TensorSplit: MainGPU: ImageGenerationAssets: PromptCachePath: PromptCacheAll:false PromptCacheRO:false Grammar: PromptStrings:[] InputStrings:[] InputToken:[] functionCallString: functionCallNameString: FunctionsConfig:{DisableNoAction:false NoActionFunctionName: NoActionDescriptionName:} SystemPrompt:}
6:21AM DBG Prompt (before templating): Hello! What is your name?
6:21AM DBG Template failed loading: failed loading a template for llama-7b-hf
6:21AM DBG Prompt (after templating): Hello! What is your name?
6:21AM DBG Model already loaded in memory: llama-7b-hf
6:21AM DBG Model 'llama-7b-hf' already loaded
[172.27.128.150]:43283  500  -  POST     /v1/chat/completions

[allenhaozi]

Deployed in k8s, the GPU has been configured, but it should not take effect

    resources:
      limits:
        cpu: "1"
        memory: 50Gi
        nvidia.com/gpu: "4"
      requests:
        cpu: "1"
        memory: 50Gi
        nvidia.com/gpu: "4"

[rozek]

I have the same problem when running LocalAI in a Docker container. The logs contain numerous lines of the form:

rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:33427: connect: connection refused"

with varying port numbers

[rozek]

FYI: the problem occurs both in local Docker builds and the ":latest" image from go-skynet

[nabbl]

Yes same here. used the latest version with GPT4All model and it just gives errors.
Same on Kubernetes and local

[Mer0me]

If it can help, my (very similar) error message :

curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
     "model": "ggml-gpt4all-j",
     "messages": [{"role": "user", "content": "How are you?"}],
     "temperature": 0.9 
   }'

{"error":{"code":500,"message":"rpc error: code = Unknown desc = unimplemented","type":""}}

[rozek]

I am currently trying to compile a previous release in order to see until when LocalAI worked without this problem.

Unfortunately, the Docker build command seems to expect the source to have been checked-out as a Git project and refuses to build from an unpacked ZIP archive...

Thus, I directly checked out v1.21.0, built a Docker image locally, ran it...and had the same problem as before.

For the records: here is what I did

git clone https://github.com/go-skynet/LocalAI.git --branch v1.21.0
cd LocalAI
docker build -t localai .
docker run --rm --name localai \
  -v "/path/to/your/local/models/folder":/build/models \
  -p 127.0.0.1:8080:8080 \
  localai

I also tried v1.20.1 and v1.20.0 - but these builds failed with "ggml.c:(.text+0x2e860): multiple definition of `clear_numa_thread_affinity'; /build/go-llama/libbinding.a(ggml.o):ggml.c:(.text+0x2e860): first defined here"

Building v1.19.2 succeeded - but it could not load my model (LLaMA 2) which makes it useless for me...

I don't have the time to check every previous version, but perhaps somebody else has...

[mudler]

did you tried running with REBUILD=true? also please attach full logs with DEBUG=true

[rozek]

Ok, so I

• checked out the latest version from GitHub,
• uncommented REBUILD=true and DEBUG=true in .env,
• rebuilt using docker build -t localai . and
• started like above using

docker run --rm --name localai \
  -v "/path/to/your/local/models/folder":/build/models \
  -p 127.0.0.1:8080:8080 \
  localai

• and tested with

curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json"  -d '{
  "model":"llama-2-13b.ggmlv3.q4_0",
  "messages": [
    {"role":"system", "content":"Perform the instructions to the best of your ability.\n"},
    {"role": "user", "content": "### Instruction: who was Joseph Weizenbaum?\n### Response:"}
  ],
  "temperature": 0.0,
  "max_tokens": 256,
  "stream": false
}'

with the same result as before. Here are the logs (mind the "skipping rebuild")

@@@@@
Skipping rebuild
@@@@@
If you are experiencing issues with the pre-compiled builds, try setting REBUILD=true
If you are still experiencing issues with the build, try setting CMAKE_ARGS and disable the instructions set as needed:
CMAKE_ARGS="-DLLAMA_F16C=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF"
see the documentation at: https://localai.io/basics/build/index.html
Note: See also https://github.com/go-skynet/LocalAI/issues/288
@@@@@
CPU info:
CPU: no AVX    found
CPU: no AVX2   found
CPU: no AVX512 found
@@@@@
3:31AM DBG no galleries to load
3:31AM INF Starting LocalAI using 4 threads, with models path: /build/models
3:31AM INF LocalAI version: v1.22.0-19-gdde12b4 (dde12b492b2da4f14d66047a42b66bff80e223af)

 ┌───────────────────────────────────────────────────┐ 
 │                   Fiber v2.48.0                   │ 
 │               http://127.0.0.1:8080               │ 
 │       (bound on host 0.0.0.0 and port 8080)       │ 
 │                                                   │ 
 │ Handlers ............ 31  Processes ........... 1 │ 
 │ Prefork ....... Disabled  PID ................ 14 │ 
 └───────────────────────────────────────────────────┘ 

rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:38673: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:41499: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:41677: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:36709: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:37449: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:33973: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:42041: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:45563: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:34759: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:42351: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:44449: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:36327: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:35585: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:45917: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:35687: connect: connection refused"
rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp 127.0.0.1:33939: connect: connection refused"

here is my .env

## Set number of threads.
## Note: prefer the number of physical cores. Overbooking the CPU degrades performance notably.
THREADS=4

## Specify a different bind address (defaults to ":8080")
ADDRESS=0.0.0.0:8080

## Default models context size
CONTEXT_SIZE=4096

## Define galleries.
## models will to install will be visible in `/models/available`
#GALLERIES=[{"name":"model-gallery", "url":"github:go-skynet/model-gallery/index.yaml"}, {"name":"huggingface","url":"github:go-skynet/model-gallery/huggingface.yaml"}]

## CORS settings
CORS=true
CORS_ALLOW_ORIGINS=*

## Default path for models
# MODELS_PATH=/models

## Enable debug mode
DEBUG=true

## Specify a build type. Available: cublas, openblas, clblas.
# BUILD_TYPE=metal

## Uncomment and set to true to enable rebuilding from source
REBUILD=true

## Enable go tags, available: stablediffusion, tts
## stablediffusion: image generation with stablediffusion
## tts: enables text-to-speech with go-piper 
## (requires REBUILD=true)
#
# GO_TAGS=stablediffusion

## Path where to store generated images
# IMAGE_PATH=/tmp

## Specify a default upload limit in MB (whisper)
# UPLOAD_LIMIT

[rozek]

here is the environment of the running container as reported by Docker (mind the "REBUILD=false")

Environment
PATH
/usr/local/cuda/bin:/go/bin:/usr/local/go/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

GOLANG_VERSION
1.20.6

GOPATH
/go

BUILD_TYPE

EXTERNAL_GRPC_BACKENDS
huggingface-embeddings:/build/extra/grpc/huggingface/huggingface.py

REBUILD
false

HEALTHCHECK_ENDPOINT
http://localhost:8080/readyz

Mounts
/build/models
/Users/andreas/rozek/AI/models/meta-ai/llama2

Port
8080/tcp
127.0.0.1:8080

[Sean-McAuliffe]

adding to this, same issues here both local docker & EKS via AL2 amd64

I can get through to /v1/models ok, but can't do anything with a model otherwise I get a timeout & various forms of:

rpc error: code = Unavailable desc = connection error: desc = "transport: Error while dialing: dial tcp [::1]:33427: connect: connection refused"

[Sean-McAuliffe]

seems it might just be an issue with the localAI image. building from scratch in a container works ok :

`
FROM public.ecr.aws/amazonlinux/amazonlinux:latest

RUN yum install git -y
RUN yum install golang -y
RUN yum group install "Development Tools" -y
RUN yum install cmake -y

RUN git clone https://github.com/go-skynet/LocalAI.git

WORKDIR /LocalAI

RUN make build

COPY . .

EXPOSE 8080

ENTRYPOINT [ "./local-ai", "--debug", "--models-path", "./models", "" ]
`

[swoh816]

@rozek @nabbl @Mer0me I had precisely the same error message as you had, so our problems may be the same. I inspected the usage of hardware resources by docker containers, and at least in my case, it was the memory limit issue. Docker Desktop (in Ubuntu 22.04) ships with a default memory limit smaller than the size of LLM (gpt4all in my case). So I set the memory limit 10GB, large enough to have gpt4all, and then it worked.

It was difficult to figure out it was the memory limit issue because the error message does not deliver it directly. Also, I don't know well about Docker, nor about LLMs, so it took some time for me to figure out the source of the problem in my machine. I think it will definitely help to include a note about increasing Docker's memory limit enough to have LLM on memory in the getting started page: https://localai.io/basics/getting_started/index.html

Note that I also uncommented REBUILD=true in .env file.
Also, increasing the memory of docker by including --memory when running container did not help either. At least in my machine, I needed to increase it in the Docker Desktop application, and it seems like a common confusion (see https://stackoverflow.com/a/44533437).

[swoh816]

@allenhaozi Also given that your debug log says it failed to load the template, I wonder if it is the issue of (1) the wrong path set to find model template, or (2) not enough memory to load template.