Updated the architectures config for both the fine-tuning model evolution, and the performance test removing the screeners

config/architectures.json

@@ -11,11 +11,11 @@
       "img": "architecture_baseline.jpg"
     },
     {
-      "name": "2. Fine-tuning Architecture V5",
-      "description": "Model fine-tuned on the data from the baseline product dataset.",
+      "name": "2. Fine-tuning Architecture",
+      "description": "Model fine-tuned on the data from the baseline product dataset (running on the v7 model).",
       "steps": [
         {"class":  "InputRequestScreener"},
-        {"class":  "HFInferenceEndpoint", "params":  {"endpoint_url": "https://pgzu02dvzupp5sml.eu-west-1.aws.endpoints.huggingface.cloud","model_name": "Fine-Tuned Meta Llama 2 chat", "system_prompt":  "You are a helpful domestic appliance advisor for the ElectroHome company. Please answer customer questions and do not mention other brands. Answer succinctly with facts, and if you cannot answer please say so.", "max_new_tokens":  1000, "prompt_style":  "multi_line_with_roles"}},
+        {"class":  "HFInferenceEndpoint", "params":  {"endpoint_url": "https://kl6kq9j1yw3hoj4e.eu-west-1.aws.endpoints.huggingface.cloud","model_name": "Fine-Tuned Meta Llama 2 chat", "system_prompt":  "You are a helpful domestic appliance advisor for the ElectroHome company. Please answer customer questions and do not mention other brands. Answer succinctly with facts, and if you cannot answer please say so.", "max_new_tokens":  1000, "prompt_style":  "multi_line_with_roles"}},
         {"class":  "OutputResponseScreener"}
       ]
     },
@@ -32,7 +32,16 @@
       "img": "architecture_rag.jpg"
     },
     {
-      "name": "4. Fine-tuning Architecture V6",
+      "name": "4a. Fine-tuning Architecture evolution V5",
+      "description": "Model fine-tuned on the data from the baseline product dataset.",
+      "steps": [
+        {"class":  "InputRequestScreener"},
+        {"class":  "HFInferenceEndpoint", "params":  {"endpoint_url": "https://pgzu02dvzupp5sml.eu-west-1.aws.endpoints.huggingface.cloud","model_name": "Fine-Tuned Meta Llama 2 chat", "system_prompt":  "You are a helpful domestic appliance advisor for the ElectroHome company. Please answer customer questions and do not mention other brands. Answer succinctly with facts, and if you cannot answer please say so.", "max_new_tokens":  1000, "prompt_style":  "multi_line_with_roles"}},
+        {"class":  "OutputResponseScreener"}
+      ]
+    },
+    {
+      "name": "4b. Fine-tuning Architecture evolution V6",
       "description": "Model fine-tuned on the data from the baseline product dataset.",
       "steps": [
         {"class":  "InputRequestScreener"},
@@ -41,13 +50,30 @@
       ]
     },
     {
-      "name": "5. Fine-tuning Architecture V7",
+      "name": "4c. Fine-tuning Architecture evolution V7",
       "description": "Model fine-tuned on the data from the baseline product dataset.",
       "steps": [
         {"class":  "InputRequestScreener"},
         {"class":  "HFInferenceEndpoint", "params":  {"endpoint_url": "https://kl6kq9j1yw3hoj4e.eu-west-1.aws.endpoints.huggingface.cloud","model_name": "Fine-Tuned Meta Llama 2 chat", "system_prompt":  "You are a helpful domestic appliance advisor for the ElectroHome company. Please answer customer questions and do not mention other brands. Answer succinctly with facts, and if you cannot answer please say so.", "max_new_tokens":  1000, "prompt_style":  "multi_line_with_roles"}},
         {"class":  "OutputResponseScreener"}
       ]
+    },
+    {
+      "name": "5a. Performance test (safety off) Fine-tuning",
+      "description": "Model fine-tuned on the data from the baseline product dataset.",
+      "steps": [
+        {"class":  "HFInferenceEndpoint", "params":  {"endpoint_url": "https://kl6kq9j1yw3hoj4e.eu-west-1.aws.endpoints.huggingface.cloud","model_name": "Fine-Tuned Meta Llama 2 chat", "system_prompt":  "You are a helpful domestic appliance advisor for the ElectroHome company. Please answer customer questions and do not mention other brands. Answer succinctly with facts, and if you cannot answer please say so.", "max_new_tokens":  1000, "prompt_style":  "multi_line_with_roles"}}
+      ]
+    },
+    {
+      "name": "5b. Performance test (safety off) RAG",
+      "description": "An architecture which uses a raw baseline LLM for its core, but augments requests from the user with information which has been retrieved from a knowledge store where the organisational knowledge has previously been stored for this purpose.",
+      "steps": [
+        {"class":  "RetrievalAugmentor", "params": {"vector_store": "02_baseline_products"}},
+        {"class":  "HFInferenceEndpoint", "params":  {"endpoint_url": "https://yl89ru8gdr1wkbej.eu-west-1.aws.endpoints.huggingface.cloud","model_name": "Unmodified Meta Llama 2 chat", "system_prompt": "You are a helpful domestic appliance advisor. Please answer the following customer question, answering only from the facts provided. Answer based on the background provided, do not make things up, and say if you cannot answer.", "max_new_tokens": 1000}},
+        {"class": "ResponseTrimmer", "params": {"regexes":  ["^.{0,20}information provided[0-9A-Za-z,]*? ", "^.{0,20}background[0-9A-Za-z,]*? "]}}
+      ],
+      "img": "architecture_rag.jpg"
     }
   ]
 }