Spaces:
Sleeping
Sleeping
Commit
·
7cb0777
1
Parent(s):
901545d
update README
Browse files
README.md
CHANGED
|
@@ -12,7 +12,7 @@ pinned: false
|
|
| 12 |
# Crypto_RAG_ChatBot
|
| 13 |
This is a cryptocurrency-focused Retrieval-Augmented Generation (RAG) app. It retrieves from your uploaded documents and added URLs, reranks results, and generates answers with a chat LLM. It can also route "price" queries to a live price tool for major coins.
|
| 14 |
|
| 15 |
-
1) OPENAI API KEY (REQUIRED; PASTE-ONLY, NEVER SAVED)
|
| 16 |
-----------------------------------------------------
|
| 17 |
- The chat model requires an OpenAI API key.
|
| 18 |
- Paste your key into the "OpenAI API Key" field in the UI.
|
|
@@ -21,7 +21,7 @@ pinned: false
|
|
| 21 |
- Restarting or refreshing the Space clears it.
|
| 22 |
- The key is used only to generate chat responses for this app.
|
| 23 |
|
| 24 |
-
2) QUICK COST ESTIMATE (GPT-4o-mini PRICING)
|
| 25 |
--------------------------------------------
|
| 26 |
Pricing used: (https://platform.openai.com/docs/pricing)
|
| 27 |
- Input: $0.15 per 1,000,000 tokens
|
|
@@ -42,13 +42,13 @@ Per-query cost:
|
|
| 42 |
|
| 43 |
Result: Approximately six-tenths of a cent ($0.0066) for the full try, well under the $0.50 budget.
|
| 44 |
|
| 45 |
-
3) FREE OPEN-SOURCE RETRIEVAL MODELS
|
| 46 |
------------------------------------
|
| 47 |
- Embeddings: sentence-transformers/all-MiniLM-L6-v2
|
| 48 |
- Reranker: cross-encoder/ms-marco-MiniLM-L-6-v2
|
| 49 |
These run locally in the Space and are free (no API costs).
|
| 50 |
|
| 51 |
-
4) UPLOADING .PDF / .TXT / .MD FILES (FOR RAG)
|
| 52 |
----------------------------------------------
|
| 53 |
- Click "Add files" and select any combination of .pdf, .txt, or .md.
|
| 54 |
Sample pdf and txt files can be found at : /data/samples -> download the
|
|
@@ -57,14 +57,14 @@ These run locally in the Space and are free (no API costs).
|
|
| 57 |
with metadata for retrieval.
|
| 58 |
- After adding files, click "Build Index" so they are included in search.
|
| 59 |
|
| 60 |
-
5) ADDING MULTIPLE URLS (FOR RAG)
|
| 61 |
---------------------------------
|
| 62 |
- Paste one URL per line in the "URLs" box and click "Add URLs".
|
| 63 |
Refer to /data/samples/Links%20sample.txt for example links
|
| 64 |
- The app fetches and parses the pages (static articles and PDFs work best).
|
| 65 |
- After adding URLs, click "Build Index" to include them in retrieval.
|
| 66 |
|
| 67 |
-
6) BUILD INDEX AND RETRIEVAL OPTIONS
|
| 68 |
------------------------------------
|
| 69 |
After you add files and/or URLs, click "Build Index". You can tune:
|
| 70 |
|
|
@@ -79,19 +79,19 @@ If results seem off:
|
|
| 79 |
- Adjust alpha (higher favors keywords; lower favors semantic similarity).
|
| 80 |
- Increase Rerank Top-K for stronger final ordering (slightly more CPU).
|
| 81 |
|
| 82 |
-
7) STREAMING RESPONSES (SELECTABLE)
|
| 83 |
---------------------------------
|
| 84 |
- Streaming ON: words appear live as the model generates.
|
| 85 |
- Streaming OFF: you receive a single final answer after generation finishes.
|
| 86 |
Toggle this with the "Streaming" checkbox.
|
| 87 |
|
| 88 |
-
8) CHAT MODEL CHOICE (FIXED FOR NOW)
|
| 89 |
------------------------------------
|
| 90 |
- The chat model is fixed in this version of the app for reliability and cost
|
| 91 |
control. If you need a different model, it can be changed in code and
|
| 92 |
redeployed; the UI currently does not expose a model selector.
|
| 93 |
|
| 94 |
-
9) LIVE PRICE SEARCH FOR MAJOR COINS + ROUTING
|
| 95 |
----------------------------------------------
|
| 96 |
- If your question looks like a price query (for example: "BTC price",
|
| 97 |
"price of ETH", "SOL price in USD"), the app routes to a tool instead of RAG:
|
|
@@ -102,14 +102,14 @@ Toggle this with the "Streaming" checkbox.
|
|
| 102 |
("price", "quote", "market cap", "ATH", and similar). If matched, it uses
|
| 103 |
the tools route; otherwise it uses the RAG route (retrieve -> rerank -> answer).
|
| 104 |
|
| 105 |
-
10) QUICK START
|
| 106 |
---------------
|
| 107 |
1. Initialize pipeline (if manual init is enabled).
|
| 108 |
2. Paste your OpenAI API key (not saved).
|
| 109 |
3. Add files and/or add URLs.
|
| 110 |
-
Sample pdf and txt files can be found at : /data/samples -> download the
|
| 111 |
files to your drive and multi select to upload
|
| 112 |
-
|
| 113 |
4. Click Build Index.
|
| 114 |
5. Ask questions. E.g. "what is Ethereum vs Solana?", "What is bitcoin strength and weakness?"
|
| 115 |
6. For prices, try queries like "ETH price", "SOL quote", "XRP price in USD".
|
|
|
|
| 12 |
# Crypto_RAG_ChatBot
|
| 13 |
This is a cryptocurrency-focused Retrieval-Augmented Generation (RAG) app. It retrieves from your uploaded documents and added URLs, reranks results, and generates answers with a chat LLM. It can also route "price" queries to a live price tool for major coins.
|
| 14 |
|
| 15 |
+
1) **OPENAI API KEY (REQUIRED; PASTE-ONLY, NEVER SAVED)**
|
| 16 |
-----------------------------------------------------
|
| 17 |
- The chat model requires an OpenAI API key.
|
| 18 |
- Paste your key into the "OpenAI API Key" field in the UI.
|
|
|
|
| 21 |
- Restarting or refreshing the Space clears it.
|
| 22 |
- The key is used only to generate chat responses for this app.
|
| 23 |
|
| 24 |
+
2) **QUICK COST ESTIMATE (GPT-4o-mini PRICING)**
|
| 25 |
--------------------------------------------
|
| 26 |
Pricing used: (https://platform.openai.com/docs/pricing)
|
| 27 |
- Input: $0.15 per 1,000,000 tokens
|
|
|
|
| 42 |
|
| 43 |
Result: Approximately six-tenths of a cent ($0.0066) for the full try, well under the $0.50 budget.
|
| 44 |
|
| 45 |
+
3) **FREE OPEN-SOURCE RETRIEVAL MODELS**
|
| 46 |
------------------------------------
|
| 47 |
- Embeddings: sentence-transformers/all-MiniLM-L6-v2
|
| 48 |
- Reranker: cross-encoder/ms-marco-MiniLM-L-6-v2
|
| 49 |
These run locally in the Space and are free (no API costs).
|
| 50 |
|
| 51 |
+
4) **UPLOADING .PDF / .TXT / .MD FILES (FOR RAG)**
|
| 52 |
----------------------------------------------
|
| 53 |
- Click "Add files" and select any combination of .pdf, .txt, or .md.
|
| 54 |
Sample pdf and txt files can be found at : /data/samples -> download the
|
|
|
|
| 57 |
with metadata for retrieval.
|
| 58 |
- After adding files, click "Build Index" so they are included in search.
|
| 59 |
|
| 60 |
+
5) **ADDING MULTIPLE URLS (FOR RAG)**
|
| 61 |
---------------------------------
|
| 62 |
- Paste one URL per line in the "URLs" box and click "Add URLs".
|
| 63 |
Refer to /data/samples/Links%20sample.txt for example links
|
| 64 |
- The app fetches and parses the pages (static articles and PDFs work best).
|
| 65 |
- After adding URLs, click "Build Index" to include them in retrieval.
|
| 66 |
|
| 67 |
+
6) **BUILD INDEX AND RETRIEVAL OPTIONS**
|
| 68 |
------------------------------------
|
| 69 |
After you add files and/or URLs, click "Build Index". You can tune:
|
| 70 |
|
|
|
|
| 79 |
- Adjust alpha (higher favors keywords; lower favors semantic similarity).
|
| 80 |
- Increase Rerank Top-K for stronger final ordering (slightly more CPU).
|
| 81 |
|
| 82 |
+
7) **STREAMING RESPONSES (SELECTABLE)**
|
| 83 |
---------------------------------
|
| 84 |
- Streaming ON: words appear live as the model generates.
|
| 85 |
- Streaming OFF: you receive a single final answer after generation finishes.
|
| 86 |
Toggle this with the "Streaming" checkbox.
|
| 87 |
|
| 88 |
+
8) **CHAT MODEL CHOICE (FIXED FOR NOW)**
|
| 89 |
------------------------------------
|
| 90 |
- The chat model is fixed in this version of the app for reliability and cost
|
| 91 |
control. If you need a different model, it can be changed in code and
|
| 92 |
redeployed; the UI currently does not expose a model selector.
|
| 93 |
|
| 94 |
+
9) **LIVE PRICE SEARCH FOR MAJOR COINS + ROUTING**
|
| 95 |
----------------------------------------------
|
| 96 |
- If your question looks like a price query (for example: "BTC price",
|
| 97 |
"price of ETH", "SOL price in USD"), the app routes to a tool instead of RAG:
|
|
|
|
| 102 |
("price", "quote", "market cap", "ATH", and similar). If matched, it uses
|
| 103 |
the tools route; otherwise it uses the RAG route (retrieve -> rerank -> answer).
|
| 104 |
|
| 105 |
+
10) **QUICK START**
|
| 106 |
---------------
|
| 107 |
1. Initialize pipeline (if manual init is enabled).
|
| 108 |
2. Paste your OpenAI API key (not saved).
|
| 109 |
3. Add files and/or add URLs.
|
| 110 |
+
- Sample pdf and txt files can be found at : /data/samples -> download the
|
| 111 |
files to your drive and multi select to upload
|
| 112 |
+
- Refer to /data/samples/Links%20sample.txt for example links
|
| 113 |
4. Click Build Index.
|
| 114 |
5. Ask questions. E.g. "what is Ethereum vs Solana?", "What is bitcoin strength and weakness?"
|
| 115 |
6. For prices, try queries like "ETH price", "SOL quote", "XRP price in USD".
|