Impostazioni per gli embeddings bge-large-en di Hugging Face? -> I bot RAG rispondono in modo sconosciuto!

Supporto
, ,
1

Si prega di consigliare.. quali sono le impostazioni migliori per consentire al modello di embedding bge-large-en di funzionare come servizio vettoriale predefinito di Discourse AI?

Ho un’istanza bge-large-en in esecuzione su AWS e so che il mio Discourse AI sta comunicando con essa (vedi test di seguito), ma l’embedding non funziona in generale (l’embedding di openAI funziona bene).

RIASSUNTO DEL PROBLEMA: i bot RAG non rispondono quando l’embedding è impostato su HF bge-large-en

ecco il modello di embedding AWS:

image
image1260×535 54 KB

ecco le impostazioni di Discourse AI:

image
image920×338 35.1 KB

image
image946×185 16.1 KB

ecco un test “Run test” di un LLM personalizzato di Discourse solo per verificare la connettività..

image
image1106×151 14.7 KB

ecco i log di bge-large-en sul lato AWS:

image
image1089×132 5.78 KB

Molte grazie!!

Ecco il log degli errori..

Job exception: can't quote Array

hostname ai-qa-ubuntu-s-1vcpu-2gb-amd-sfo3-01-app
process_id 1165935
application_version f9192835a7e4d2067c3d1844f43f9e7b69de39e7
current_db default
current_hostname ai-qa.net
job Jobs::CreateAiReply
problem_db default
time 7:22 pm
opts post_id 618
--- --- --- ---
--- ---
bot_user_id -1208
persona_id 5
current_site_id default


Backtrace

/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/connection_adapters/abstract/quoting.rb:25:in `quote'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/connection_adapters/postgresql/quoting.rb:69:in `quote'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/connection_adapters/abstract/quoting.rb:51:in `quote_bound_value'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:193:in `block in quote_bound_value'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:193:in `map!'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:193:in `quote_bound_value'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:171:in `replace_bind_variable'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:180:in `block in replace_named_bind_variables'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:176:in `gsub'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:176:in `replace_named_bind_variables'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activerecord-7.0.8.1/lib/active_record/sanitization.rb:128:in `sanitize_sql_array'
/var/www/discourse/lib/mini_sql_multisite_connection.rb:21:in `public_send'
/var/www/discourse/lib/mini_sql_multisite_connection.rb:21:in `encode'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/mini_sql-1.5.0/lib/mini_sql/connection.rb:64:in `to_sql'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/mini_sql-1.5.0/lib/mini_sql/postgres/connection.rb:202:in `run'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/mini_sql-1.5.0/lib/mini_sql/active_record_postgres/connection.rb:38:in `block in run'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/mini_sql-1.5.0/lib/mini_sql/active_record_postgres/connection.rb:34:in `block in with_lock'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activesupport-7.0.8.1/lib/active_support/concurrency/load_interlock_aware_monitor.rb:25:in `handle_interrupt'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activesupport-7.0.8.1/lib/active_support/concurrency/load_interlock_aware_monitor.rb:25:in `block in synchronize'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activesupport-7.0.8.1/lib/active_support/concurrency/load_interlock_aware_monitor.rb:21:in `handle_interrupt'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/activesupport-7.0.8.1/lib/active_support/concurrency/load_interlock_aware_monitor.rb:21:in `synchronize'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/mini_sql-1.5.0/lib/mini_sql/active_record_postgres/connection.rb:34:in `with_lock'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/mini_sql-1.5.0/lib/mini_sql/active_record_postgres/connection.rb:38:in `run'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/mini_sql-1.5.0/lib/mini_sql/postgres/connection.rb:99:in `query'
/var/www/discourse/plugins/discourse-ai/lib/embeddings/vector_representations/base.rb:272:in `asymmetric_rag_fragment_similarity_search'
/var/www/discourse/plugins/discourse-ai/lib/ai_bot/personas/persona.rb:286:in `rag_fragments_prompt'
/var/www/discourse/plugins/discourse-ai/lib/ai_bot/personas/persona.rb:156:in `craft_prompt'
/var/www/discourse/plugins/discourse-ai/lib/ai_bot/bot.rb:54:in `reply'
/var/www/discourse/plugins/discourse-ai/lib/ai_bot/playground.rb:424:in `reply_to'
/var/www/discourse/plugins/discourse-ai/app/jobs/regular/create_ai_reply.rb:18:in `execute'
/var/www/discourse/app/jobs/base.rb:305:in `block (2 levels) in perform'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/rails_multisite-6.0.0/lib/rails_multisite/connection_management/null_instance.rb:49:in `with_connection'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/rails_multisite-6.0.0/lib/rails_multisite/connection_management.rb:21:in `with_connection'
/var/www/discourse/app/jobs/base.rb:292:in `block in perform'
/var/www/discourse/app/jobs/base.rb:288:in `each'
/var/www/discourse/app/jobs/base.rb:288:in `perform'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:202:in `execute_job'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:170:in `block (2 levels) in process'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/middleware/chain.rb:177:in `block in invoke'
/var/www/discourse/lib/sidekiq/pausable.rb:132:in `call'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/middleware/chain.rb:179:in `block in invoke'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/middleware/chain.rb:182:in `invoke'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:169:in `block in process'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:136:in `block (6 levels) in dispatch'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/job_retry.rb:113:in `local'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:135:in `block (5 levels) in dispatch'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq.rb:44:in `block in <module:Sidekiq>'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:131:in `block (4 levels) in dispatch'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:263:in `stats'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:126:in `block (3 levels) in dispatch'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/job_logger.rb:13:in `call'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:125:in `block (2 levels) in dispatch'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/job_retry.rb:80:in `global'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:124:in `block in dispatch'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/job_logger.rb:39:in `prepare'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:123:in `dispatch'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:168:in `process'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:78:in `process_one'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/processor.rb:68:in `run'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/component.rb:8:in `watchdog'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/sidekiq-6.5.12/lib/sidekiq/component.rb:17:in `block in safe_thread'
2 Mi Piace
3

Grazie per aver segnalato, ci daremo un’occhiata!

2 Mi Piace
4

Qual è l’output dell’esecuzione dei seguenti comandi in una console Rails?

strategy = DiscourseAi::Embeddings::Strategies::Truncation.new
vector_rep = DiscourseAi::Embeddings::VectorRepresentations::Base.current_representation(strategy)
vector_rep.vector_from("test")

Inoltre, la nostra API è progettata per funzionare contro qualcuno che esegue GitHub - huggingface/text-embeddings-inference: A blazing fast inference solution for text embeddings models autonomamente, come da documentazione, quindi è possibile che non funzioni contro la versione ospitata.

Se fornisci il backtrace, possiamo studiare come farlo funzionare.

2 Mi Piace
5

@Falco

ecco cosa è successo quando ho eseguito il codice di test (con bge-large-en in esecuzione su un’istanza di endpoint dedicata AWS configurata come modello di embedding)

image
image833×121 9.63 KB

root@studyqa-ubuntu-s-1vcpu-2gb-amd-sfo3-01-app:/var/www/discourse# rails c

[1] pry(main)> strategy = DiscourseAi::Embeddings::Strategies::Truncation.new

puts "Strategy initialized"

vector_rep = DiscourseAi::Embeddings::VectorRepresentations::Base.current_representation(strategy)

puts "Vector representation obtained"

vector = vector_rep.vector_from("test")

[1] pry(main)> strategy = DiscourseAi::Embeddings::Strategies::Truncation.new

puts "Strategy initialized"

vector_rep = DiscourseAi::Embeddings::VectorRepresentations::Base.current_representation(strategy)

puts "Vector representation obtained"

vector = vector_rep.vector_from("test")

puts "Vector generated"

puts vector.inspect

Strategy initialized

Vector representation obtained

Vector generated

[:embeddings, [-0.0020444912370294333, 0.008787356317043304, -0.010865539312362671, 0.01865551434457302, -0.02099628746509552, -0.009864491410553455, -0.0011329081607982516, 0.02949545904994011, 0.027839021757245064, 0.043966952711343765, 0.0406080037355423, 0.0016647017328068614, 0.007204003632068634, -0.03770752251148224, -0.025242917239665985, -0.0015279072104021907, -0.02805529721081257, -0.020901955664157867, -0.029206447303295135, -0.006209365092217922, -0.02105099707841873,

ecc.

sembra che stia raggiungendo bge-large-en in aws:

- 2024-05-29T13:57:34.609+00:00 Batches: 0%| | 0/1 [00:00<?, ?it/s] Batches: 100%|██████████| 1/1 [00:00<00:00, 4.80it/s] Batches: 100%|██████████| 1/1 [00:00<00:00, 4.79it/s]

• 2024/05/29 09:57:34
INFO | POST / | Duration: 212.84 ms


- 2024-05-29T13:57:53.806+00:00 Batches: 0%| | 0/1 [00:00<?, ?it/s] Batches: 100%|██████████| 1/1 [00:01<00:00, 1.97s/it] Batches: 100%|██████████| 1/1 [00:01<00:00, 1.97s/it]

• 2024/05/29 09:57:53
INFO | POST / | Duration: 1978.36 ms
1 Mi Piace
6

Quindi sembra che funzioni perfettamente?

Forse il problema è il re-ranker? Puoi annullare l’impostazione di ai_hugging_face_tei_reranker_endpoint e verificare se RAG funziona?

7

reranker è stato disattivato.. nessun embedding ancora.. sto ricevendo questo messaggio su entrambe le estremità:

Test di esecuzione Discourse LLM:

Il tentativo di contattare il modello ha restituito questo errore: {“error”:“Il corpo deve fornire una chiave ‘inputs’, ricevuto: b’{"model":"bge-large-en","temperature":0.7,"messages":[{"role":"system","content":"You are a helpful bot"},{"role":"user","content":"How much is 1 + 1?"}],"max_tokens":1009}'”}

Log bge-large-en

• 2024/05/29 13:40:03

ERRORE | Il corpo deve fornire una chiave ‘inputs’, ricevuto: b’{"model":"bge-large-en","temperature":0.7,"messages":[{"role":"system","content":"You are a helpful bot"},{"role":"user","content":"How much is 1 + 1?"}],"max_tokens":1009}’

discourse b1b218aa99
discourse-ai d812ecf5

8

Questo non è il modo in cui dovremmo testare gli embedding :slight_smile: quello è un test LLM non un test di modello di embedding che si aspetterebbe numeri in risposta. L’interfaccia utente LLM non è dove lo aggiungeresti, avremmo bisogno di un’interfaccia utente di Embedding che non abbiamo ancora. I modelli di embedding sono configurati solo nelle impostazioni del sito.

9

Sì. Ha senso.

(Ho cercato di far notare che stavo usando solo il test eseguito dall’LLM per confermare la “connettività” (vedi sotto)! Avrei dovuto renderlo più chiaro.)

image
image715×305 35.5 KB

1 Mi Piace