FIX: Don't echo prompt back on HF/TGI (#338)

* FIX: Don't echo prompt back on HF/TGI

* teeeeests

lib/completions/endpoints/hugging_face.rb

@@ -33,6 +33,7 @@ module DiscourseAi
               token_limit = SiteSetting.ai_hugging_face_token_limit || 4_000
 
               payload[:parameters][:max_new_tokens] = token_limit - prompt_size(prompt)
+              payload[:parameters][:return_full_text] = false
 
               payload[:stream] = true if @streaming_mode
             end

lib/inference/hugging_face_text_generation.rb

@@ -40,6 +40,7 @@ module ::DiscourseAi
         parameters[:max_new_tokens] = token_limit - prompt_size
         parameters[:temperature] = temperature if temperature
         parameters[:repetition_penalty] = repetition_penalty if repetition_penalty
+        parameters[:return_full_text] = false
 
         payload[:stream] = true if block_given?
 

spec/lib/completions/endpoints/hugging_face_spec.rb

@@ -18,6 +18,7 @@ RSpec.describe DiscourseAi::Completions::Endpoints::HuggingFace do
       .tap do |payload|
         payload[:parameters][:max_new_tokens] = (SiteSetting.ai_hugging_face_token_limit || 4_000) -
           model.prompt_size(prompt)
+        payload[:parameters][:return_full_text] = false
       end
       .to_json
   end
@@ -29,6 +30,7 @@ RSpec.describe DiscourseAi::Completions::Endpoints::HuggingFace do
         payload[:parameters][:max_new_tokens] = (SiteSetting.ai_hugging_face_token_limit || 4_000) -
           model.prompt_size(prompt)
         payload[:stream] = true
+        payload[:parameters][:return_full_text] = false
       end
       .to_json
   end