Turn llm into an NFT

whoami - a new endpoint that returns principal of caller in text format http_request endpoint handle preflight request (OPTIONS) charles NFTCollection nft_mint ; http_request Cleanup and fix all tests
icppWorld · Dec 23, 2023 · 8c0f7a5 · 8c0f7a5
1 parent f3d7b2d
commit 8c0f7a5
Show file tree

Hide file tree

Showing 25 changed files with 25,283 additions and 60 deletions.
diff --git a/.clang-format b/.clang-format
@@ -105,7 +105,7 @@ PenaltyExcessCharacter: 1000000
 PenaltyReturnTypeOnItsOwnLine: 60
 PointerAlignment: Right
 ReflowComments: false
-SortIncludes: true
+SortIncludes: false
 SortUsingDeclarations: true
 SpaceAfterCStyleCast: false
 SpaceAfterLogicalNot: false

diff --git a/icpp_llama2/README.md b/icpp_llama2/README.md
@@ -44,10 +44,31 @@
 
 ![icpp_llama2_without_limits](../assets/icpp_llama2_without_limits.png)
 
+- To deploy all the LLM canisters and upload their models + tokenizers:
+  ```bash
+  # Build WASM & deploy all the canisters listed in dfx.json
+  icpp build-wasm
+  dfx deploy
+
+  # Upload the models & tokenizers to the canisters
+  # Notes:
+  # (-) The storiesXX.bin files are not included in the github repo
+  # (-) See `Get a model checkpoint` above
+  # (-) The default tokenizer.bin is included in the github repo
+  # (-) See `stories260k` below how to build the tok512.bin for the stories260K model
+  #
+  python -m scripts.upload --canister llama2_260K --model stories260K/stories260K.bin --tokenizer stories260K/tok512.bin
+
+  python -m scripts.upload --canister llama2 --model models/stories15M.bin --tokenizer tokenizer/tokenizer.bin
+
+  python -m scripts.upload --canister llama2_42M --model models/stories42M.bin --tokenizer tokenizer/tokenizer.bin
+
+  python -m scripts.upload --canister llama2_110M --model models/stories110M.bin --tokenizer tokenizer/tokenizer.bin
+  ```
 
 # stories260K
 
-The default model is`stories15M.bin`, with `tokenizer.bin`, which contains the default llama2 tokenizer using 32000 tokens. 
+The default model is `stories15M.bin`, with `tokenizer.bin`, which contains the default llama2 tokenizer using 32000 tokens. 
 
 For testing, it is nice to be able to work with a smaller model & tokenizer:
 - Download the model & tokenizer from [huggingface stories260K](https://huggingface.co/karpathy/tinyllamas/tree/main/stories260K) and store them in:
@@ -61,16 +82,35 @@ For testing, it is nice to be able to work with a smaller model & tokenizer:
   ```
 - Upload the model & tokenizer:
   ```bash
-  python -m scripts.upload --model stories260K/stories260K.bin --tokenizer stories260K/tok512.bin
+  python -m scripts.upload --canister llama2_260K --model stories260K/stories260K.bin --tokenizer stories260K/tok512.bin
   ```
 - Inference is now possible with many more tokens before hitting the instruction limit, but off course, the stories are not as good:
   ```bash
-  $ dfx canister call llama2 inference '(record {prompt = "Lilly went swimming yesterday  " : text; steps = 100 : nat64; temperature = 0.9 : float32; topp = 0.9 : float32; rng_seed = 0 : nat64;})'
+  # Create a new chat
+  $ dfx canister call llama2_260K new_chat '()'
+
+  # Start the chat by providing the starting prompt
+  $ dfx canister call llama2_260K inference '(record {prompt = "Lilly went swimming yesterday  " : text; steps = 100 : nat64; temperature = 0.9 : float32; topp = 0.9 : float32; rng_seed = 0 : nat64;})'
   (
     variant {
       ok = "Lilly went swimming yesterday  order. She had a great eyes that was closed. One day, she asked her mom why the cloud was close to the pond. \n\"Mommy, I will take clothes away,\" Lila said. \"Th\n"
     },
   )
+
+  # Continue the current chat by calling again, with an empty prompt
+  $ dfx canister call llama2_260K inference '(record {prompt = "" : text; steps = 100 : nat64; temperature = 0.9 : float32; topp = 0.9 : float32; rng_seed = 0 : nat64;})'
+  (
+    variant {
+      ok = "eone replace it.\nThe fox agreed to go as fast as they set on the other birds. They searched, and it didn\'t give up. They started to scared the bird. The forest was so careful and jumped up."
+    },
+  )
+
+  # Retrieve your full story, by calling with curl, passing the principal by which the LLM knows you in the body
+  $ dfx canister call llama2_260K whoami
+  '("<your-principal>")'
+
+  $ curl -X GET -d '{"principal":"<your-principal>"}' http://localhost:$(dfx info webserver-port)?canisterId=$(dfx canister id llama2_260K)
+
   ```
 
 # Fine tuning

diff --git a/icpp_llama2/canister_ids.json b/icpp_llama2/canister_ids.json
@@ -1,4 +1,7 @@
 {
+  "charles": {
+    "ic": "lkh5o-3yaaa-aaaag-acguq-cai"
+  },
   "llama2": {
     "ic": "4c4bn-daaaa-aaaag-abvcq-cai"
   },

diff --git a/icpp_llama2/demo.ps1 b/icpp_llama2/demo.ps1
@@ -56,8 +56,9 @@ wsl --% dfx canister call llama2 health
 #######################################################################
 Write-Host " "
 Write-Host "--------------------------------------------------"
-Write-Host "Upload the model & tokenizer"
-python -m scripts.upload
+Write-Host "Upload the model & tokenizer for 260K and 15M"
+python -m scripts.upload --canister llama2_260K --model stories260K/stories260K.bin --tokenizer stories260K/tok512.bin
+python -m scripts.upload --canister llama2 --model models/stories15M.bin --tokenizer tokenizer/tokenizer.bin
 
 #######################################################################
 Write-Host " "

diff --git a/icpp_llama2/demo.sh b/icpp_llama2/demo.sh
@@ -35,8 +35,9 @@ dfx canister call llama2 health
 #######################################################################
 echo " "
 echo "--------------------------------------------------"
-echo "Upload the model & tokenizer"
-python -m scripts.upload
+echo "Upload the model & tokenizer for 260K and 15M"
+python -m scripts.upload --canister llama2_260K --model stories260K/stories260K.bin --tokenizer stories260K/tok512.bin
+python -m scripts.upload --canister llama2 --model models/stories15M.bin --tokenizer tokenizer/tokenizer.bin
 
 #######################################################################
 echo " "

diff --git a/icpp_llama2/dfx.json b/icpp_llama2/dfx.json
@@ -20,6 +20,11 @@
       "type": "custom",
       "candid": "src/llama2.did",
       "wasm": "build/llama2.wasm"
+    },
+    "charles": {
+      "type": "custom",
+      "candid": "src/llama2.did",
+      "wasm": "build/llama2.wasm"
     }
   },
   "defaults": {

diff --git a/icpp_llama2/icpp.toml b/icpp_llama2/icpp.toml
@@ -1,18 +1,15 @@
 [build-wasm]
-canister          = "llama2"
-did_path          = "src/llama2.did"
-cpp_paths         = ["src/*.cpp"]
-cpp_header_paths  = ["src/*.h"]
-cpp_compile_flags = []
-cpp_link_flags    = []
-c_paths           = ["src/run.c"]
-c_header_paths    = ["src/run.h"]
-c_compile_flags   = []
+canister = "llama2"
+did_path = "src/llama2.did"
+cpp_paths = ["src/*.cpp"]
+cpp_include_dirs = ["src/vendors/*"]
+cpp_compile_flags = ["-D JSON_HAS_FILESYSTEM=0"]
+cpp_link_flags = []
+c_paths = ["src/run.c"]
+c_compile_flags = []
 [build-native]
-cpp_paths         = ["native/main.cpp"]
-cpp_header_paths  = ["native/main.h"]
+cpp_paths = ["native/main.cpp"]
 cpp_compile_flags = []
-cpp_link_flags    = []
-c_paths           = []
-c_header_paths    = []
-c_compile_flags   = []
+cpp_link_flags = []
+c_paths = []
+c_compile_flags = []
diff --git a/icpp_llama2/native/main.cpp b/icpp_llama2/native/main.cpp
@@ -21,8 +21,10 @@
 
 #include "../src/canister.h"
 #include "../src/chats.h"
+#include "../src/http.h"
 #include "../src/inference.h"
 #include "../src/initialize.h"
+#include "../src/nfts.h"
 #include "../src/upload.h"
 #include "../src/users.h"
 
@@ -66,7 +68,7 @@ int main() {
   bool silent_on_trap = true;
 
   // The model & tokenizer to use
-  int model_to_use = 2; // 1=260K, 2=15M, 3=42M, 4=110M (TinyStories)
+  int model_to_use = 1; // 1=260K, 2=15M, 3=42M, 4=110M (TinyStories)
 
   std::string model_path;
   std::string tokenizer_path;
@@ -177,6 +179,12 @@ int main() {
   mockIC.run_test("ready", ready, "4449444c0000", "4449444c00017e00",
                   silent_on_trap, anonymous_principal);
 
+  // '()' -> '("expmt-gtxsw-inftj-ttabj-qhp5s-nozup-n3bbo-k7zvn-dg4he-knac3-lae")'
+  mockIC.run_test(
+      "whoami", whoami, "4449444c0000",
+      "4449444c0001713f6578706d742d67747873772d696e66746a2d747461626a2d71687035732d6e6f7a75702d6e3362626f2d6b377a766e2d64673468652d6b6e6163332d6c6165",
+      silent_on_trap, my_principal);
+
   // '()' -> '(variant { err = 401 : nat16 })'
   // Call with non owner principal must be denied
   mockIC.run_test("reset_model err", reset_model, "4449444c0000",
@@ -461,7 +469,7 @@ int main() {
     // '()' -> '(variant { ok = 200 : nat16 })'
     mockIC.run_test("new_chat", new_chat, "4449444c0000",
                     "4449444c016b019cc2017a010000c800", silent_on_trap,
-                    my_principal);
+                    your_principal);
 
     // Loop to create 1000 token long story, 10 tokens at a time
     // With temperature=0.0: greedy argmax sampling -> the story will be the same every time
@@ -484,7 +492,7 @@ int main() {
 
       std::string candid_out;
       mockIC.run_test("inference 0b", inference, candid_in, "", silent_on_trap,
-                      my_principal, &candid_out);
+                      your_principal, &candid_out);
 
       std::string err_text;
       CandidTypeVariant v_out;
@@ -505,6 +513,77 @@ int main() {
     std::cout << story;
   }
 
+  // ------------------------------------------------------------------------
+  // Mint current conversation of 'your_principal'
+
+  // Verify it traps when not owner of the canister
+  // '(record {nft_owner = principal "2ibo7-dia"})' -> trap
+  mockIC.run_trap_test("nft_mint trap test", nft_mint,
+                       "4449444c016c01b08bb1cf07680100010100", silent_on_trap,
+                       your_principal);
+
+  // '(record {nft_owner = principal "2ibo7-dia"})'
+  // ->
+  // (
+  //   record {
+  //     51_777_086 = 1 : nat;                        // nft_id
+  //     2_045_527_472 = principal "2ibo7-dia";       // nft_owner
+  //     3_314_270_578 = "Once upon a time, ...";     // nft_story
+  //   },
+  // )
+  expected_response = "-to-do-";
+  if (model_to_use == 1) {
+    expected_response =
+        "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";
+  } else if (model_to_use == 2) {
+  } else if (model_to_use == 3) {
+  } else if (model_to_use == 4) {
+  }
+  mockIC.run_test("nft_mint", nft_mint, "4449444c016c01b08bb1cf07680100010100",
+                  expected_response, silent_on_trap, my_principal);
+
+  // ------------------------------------------------------------------------
+  // Call to the http_request endpoint, to get the story of nft with id "1"
+  // (
+  //   record {                                       // IC_HttpRequest
+  //     5_843_823 = "/api/nft/1";                    // url
+  //     156_956_385 = "GET";                         // method
+  //     1_092_319_906 = vec {};                      // body
+  //     1_661_489_734 = vec { record {...} };        // headers
+  //     1_661_892_784 = opt (2 : nat16);             // certificate_version
+  //   },
+  // )
+  // I encoded the anonimized version of an actual request to MAINNET:
+  // '(record { 5_843_823 = "/api/nft/1"; 156_956_385 = "GET"; 1_092_319_906 = blob "{}"; 1_661_489_734 = vec { record { "host"; "xxxxx-xxxxx-xxxxx-xxxxx-cai.icp0.io";}; record { "x-real-ip"; "xx.xx.xxx.xxx";}; record { "x-forwarded-for"; "xx.xx.xxx.xxx";}; record { "x-forwarded-proto"; "https";}; record { "x-request-id"; "15aad0a6-e422-26e5-8992-c3fe477ffedd";}; record { "x-icx-require-certification"; "1";}; record { "pragma"; "no-cache";}; record { "cache-control"; "no-cache";}; record { "sec-ch-ua"; "\"Not_A Brand\";v=\"8\", \"Chromium\";v=\"120\", \"Brave\";v=\"120\"";}; record { "sec-ch-ua-mobile"; "?0";}; record { "user-agent"; "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36";}; record { "sec-ch-ua-platform"; "\"Linux\"";}; record { "accept"; "*/*";}; record { "sec-gpc"; "1";}; record { "origin"; "null";}; record { "sec-fetch-site"; "cross-site";}; record { "sec-fetch-mode"; "cors";}; record { "sec-fetch-dest"; "empty";}; record { "accept-encoding"; "gzip, deflate, br, identity";}; record { "accept-language"; "en-US,en;q=0.9";};}; 1_661_892_784 = opt (2 : nat16); }, )'
+  //
+  // Note: Encoding this gives an error on the empty vec, not sure yet why:
+  // '(record { 5_843_823 = "/api/nft/1"; 156_956_385 = "GET"; 1_092_319_906 = vec {}; 1_661_489_734 = vec { record { "host"; "xxxxx-xxxxx-xxxxx-xxxxx-cai.icp0.io";}; record { "x-real-ip"; "xx.xx.xxx.xxx";}; record { "x-forwarded-for"; "xx.xx.xxx.xxx";}; record { "x-forwarded-proto"; "https";}; record { "x-request-id"; "15aad0a6-e422-26e5-8992-c3fe477ffedd";}; record { "x-icx-require-certification"; "1";}; record { "pragma"; "no-cache";}; record { "cache-control"; "no-cache";}; record { "sec-ch-ua"; "\"Not_A Brand\";v=\"8\", \"Chromium\";v=\"120\", \"Brave\";v=\"120\"";}; record { "sec-ch-ua-mobile"; "?0";}; record { "user-agent"; "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36";}; record { "sec-ch-ua-platform"; "\"Linux\"";}; record { "accept"; "*/*";}; record { "sec-gpc"; "1";}; record { "origin"; "null";}; record { "sec-fetch-site"; "cross-site";}; record { "sec-fetch-mode"; "cors";}; record { "sec-fetch-dest"; "empty";}; record { "accept-encoding"; "gzip, deflate, br, identity";}; record { "accept-language"; "en-US,en;q=0.9";};}; 1_661_892_784 = opt (2 : nat16); }, )'
+  //
+  // -> Returns an IC_HttpResponse
+  // (
+  //   record {                                                                     // IC_HttpResponse
+  //     1_092_319_906 = blob "{\22story\22:\22Once upon a time, ...\5c\22\22}";    // body
+  //     1_661_489_734 = vec {                                                      // headers
+  //       record { "Content-Type"; "application/json" };
+  //       record { "Content-Length"; "803" };
+  //     };
+  //     1_664_201_884 = opt false;                                                 // upgrade
+  //     3_475_804_314 = 200 : nat16;                                               // status_code
+  //   },
+  // )
+  expected_response = "-to-do-";
+  if (model_to_use == 1) {
+    expected_response =
+        "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";
+  } else if (model_to_use == 2) {
+  } else if (model_to_use == 3) {
+  } else if (model_to_use == 4) {
+  }
+  mockIC.run_test(
+      "http_request anonimized with blob body", http_request,
+      "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",
+      expected_response, silent_on_trap, my_principal);
+
   // -----------------------------------------------------------------------------------------
   // A new chat
   // '()' -> '(variant { ok = 200 : nat16 })'