! Unexpected content type "text/plain".

[Edouard-Legoupil]

Hello,

First congrat for the documentation! Really extensive and nicely made! Always an engaging sign!

Testing package - installed with ollama running on linux...

ping_ollama() ▶ Ollama is running at http://localhost:11434!

When I try any query: query("why is the sky blue?")

I get this error:

httr2::resp_body_json(httr2::req_perform(httr2::req_error(httr2::req_body_json(httr2::req_url_path_append(httr2::request(server), …:

! Unexpected content type "text/plain".

More below

.Last.error <callr_error/rlib_error_3_0/rlib_error/error> Error: ! in callr subprocess.

---

Backtrace:

1. rollama::query("why is the sky blue?")
2. rollama:::build_req(model = model, msg = msg, server = server, images = images, …
3. rp$get_result()
4. callr:::rp_get_result(self, private)
5. callr:::get_result(out, private$options)

6. callr:::throw(callr_remote_error(remerr, output), parent = fix_msg(remerr[[3]]))

   Subprocess backtrace: . `

Any hints?

Thanks! Edouard

[JBGruber]

Hi,

thanks for filing the issue. It does not really make sense to me at this point. But can you please try the follwing and post the output:

reprex::reprex({
  library(rollama)
  options("rollama_verbose" = FALSE)
  query("why is the sky blue?")
  system2("curl", "-V", stdout = TRUE)
}, session_info = TRUE)

FYI: options("rollama_verbose" = FALSE) gives you better error messages because callr is not involved anymore (I realise this is confusing, but do not know at the moment how to do it better).

[JBGruber]

Very odd, I just encountered the same issue. But it has nothing to do with rollama. You can test that by using:

curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
  "prompt": "What color is the sky at different times of the day? Respond using JSON",
  "format": "json",
  "stream": false
}'

In my case this suddenly returns curl: (52) Empty reply from server.

And if you get an empty reply like me, also try:

docker exec -it ollama ollama run llama2
# or if you run ollama without docker
ollama run llama2

Both fail for me at the moment. Interestingly, other calls to the API, for example, to list available modesl, succeed.

Update:

It seems another process was hogging my vRAM. By turning off the GPU, you can force the model to use your system memory (RAM). So this works for me:

query("why is the sky blue?", model_params = list(num_gpu = 0))

This could be solved much better upstream! There are no logs or errors indicating that this is a OOM error.

[Edouard-Legoupil]

On # 1 -

library(rollama) options("rollama_verbose" = FALSE) query("why is the sky blue?")

> Error in httr2::resp_body_json():

> ! Unexpected content type "text/plain".

> • Expecting type "application/json" or suffix "json".

> Backtrace:

> ▆

> 1. └─rollama::query("why is the sky blue?")

> 2. └─rollama:::build_req(...)

> 3. └─rollama:::make_req(req_data, server, "/api/chat")

> 4. └─httr2::resp_body_json(...)

> 5. └─httr2::resp_check_content_type(...)

> 6. └─httr2:::check_content_type(...)

> 7. └─cli::cli_abort(...)

> 8. └─rlang::abort(...)

system2("curl", "-V", stdout = TRUE)

> [1] "curl 7.68.0 (x86_64-pc-linux-gnu) libcurl/7.68.0 OpenSSL/1.1.1f zlib/1.2.11 brotli/1.0.7 libidn2/2.2.0 libpsl/0.21.0 (+libidn2/2.2.0) libssh/0.9.3/openssl/zlib nghttp2/1.40.0 librtmp/2.3"

> [2] "Release-Date: 2020-01-08"

> [3] "Protocols: dict file ftp ftps gopher http https imap imaps ldap ldaps pop3 pop3s rtmp rtsp scp sftp smb smbs smtp smtps telnet tftp "

> [4] "Features: AsynchDNS brotli GSS-API HTTP2 HTTPS-proxy IDN IPv6 Kerberos Largefile libz NTLM NTLM_WB PSL SPNEGO SSL TLS-SRP UnixSockets"

[Edouard-Legoupil]

Tried also

FYI - I am using this - https://github.com/docker/genai-stack to launch my ollama

options("rollama_verbose" = FALSE) ping_ollama() ▶ Ollama is running at http://localhost:11434! query("why is the sky blue?", model_params = list(num_gpu = 0)) Error in httr2::resp_body_json(): ! Unexpected content type "text/plain". • Expecting type "application/json" or suffix "json". Run rlang::last_trace() to see where the error occurred.

[JBGruber]

I read somewhere that Ollama does not work with old versions of curl (I can't find the source). The one I'm using is almost 4 years newer than yours (curl 8.5.0 (x86_64-pc-linux-gnu), Release-Date: 2023-12-06). I would suggest you update that. Then I would be interested in seeing what this does (in terminal, not R):

curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
  "prompt": "What color is the sky at different times of the day? Respond using JSON",
  "format": "json",
  "stream": false
}'

If this also fails after updating curl, this is an Ollama problem that I can't really help with.

[Edouard-Legoupil]

yes - had to do a big LTS update...! Thanks! now it works!!