From 0cd68badfcc924ad3da43c9cb5b5c1babbee888a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?David=20Lanzend=C3=B6rfer?= <leviathan@libresilicon.com>
Date: Fri, 28 Jun 2024 00:07:00 +0100
Subject: [PATCH] Prepare testing of training cycles
---
.gitlab-ci.yml | 11 ++++++++
firmware/firmware.c | 43 +++++++++++++++++++++++++++-
firmware/rnn.c | 68 +++++++++++++--------------------------------
src/py/tty3.py | 33 ++++++++++++++++++++++
4 files changed, 105 insertions(+), 50 deletions(-)
create mode 100644 src/py/tty3.py
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 533b16f..83e3d8e 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -27,6 +27,17 @@ write_test:
paths:
- result
+train_test:
+ stage: build
+ script:
+ - mkdir -p result
+ - make
+ - ./testbench_ecp5_minifpga.bin &
+ - sleep 1s && python3 src/py/tty3.py
+ artifacts:
+ paths:
+ - result
+
#json:
# stage: build
# script:
diff --git a/firmware/firmware.c b/firmware/firmware.c
index 0b0af18..cccdc66 100644
--- a/firmware/firmware.c
+++ b/firmware/firmware.c
@@ -23,6 +23,7 @@
#define MEM_TOTAL 0x20000
#define MSGBUF 40
+#define MAX_NUM_TOKENS 20
// a pointer to this is a null pointer, but the compiler does not
// know that because "sram" is a linker symbol from sections.lds.
@@ -66,7 +67,11 @@ void main()
// Write mode
WRITE,
WRITE_LAYER,
- WRITE_LAYER_WEIGHTS
+ WRITE_LAYER_WEIGHTS,
+ // Training the network
+ TRAIN,
+ TRAIN_STORE_TOKENS,
+ TRAIN_PROCESS
} command_mode;
command_mode = START;
@@ -75,9 +80,15 @@ void main()
uint32_t current_max_num_values;
uint32_t current_num_neurons;
+ // Write params
uint32_t value_write_counter;
int new_value;
+ // Training
+ uint32_t new_token;
+ uint32_t token_series[MAX_NUM_TOKENS];
+ int token_counter;
+
while(true) {
syn(); // Booted
@@ -115,6 +126,12 @@ void main()
else if(!strcmp(msg,"WRITE")) {
command_mode = WRITE;
}
+ else if(!strcmp(msg,"RESET")) {
+ reset_network();
+ }
+ else if(!strcmp(msg,"TRAIN")) {
+ command_mode = TRAIN;
+ }
break;
case INIT:
@@ -232,6 +249,30 @@ void main()
value_write_counter++;
}
break;
+
+ case TRAIN:
+ if(!strcmp(msg,"TOKENS")) {
+ command_mode = TRAIN_STORE_TOKENS;
+ response = "OK";
+ token_counter = 0;
+ }
+ else if(!strcmp(msg,"RUN")) {
+ command_mode = TRAIN_PROCESS;
+ response = "OK";
+ }
+ break;
+
+ case TRAIN_STORE_TOKENS:
+ if(token_counter<MAX_NUM_TOKENS) {
+ new_token = atoi(numstr,10);
+ token_series[token_counter] = new_token;
+ token_counter++;
+ response = "OK";
+ } else {
+ response = "END";
+ }
+ break;
+
}
write_response(response);
}
diff --git a/firmware/rnn.c b/firmware/rnn.c
index 06e015b..af34057 100644
--- a/firmware/rnn.c
+++ b/firmware/rnn.c
@@ -394,62 +394,32 @@ void set_bias_values(int bias)
}
}
-#if 0
+void run_training()
+{
int dw = 200000000;
int bias = 0;
int last_val;
uint32_t train_mask;
- /*
-
- citoa(27017, numstr, 10);
- print("Input value: ");
- print(numstr);
- ack();
-
- last_val = predict_next_token(27017);
-
- citoa(last_val, numstr, 10);
- print("Output value: ");
- print(numstr);
- ack();
-
- leds(8);*/
- /* Compare values
- * The phrase "Taxation is theft"
- * tokenized with the GPT2 tokenizer
- * is tensor([[27017, 341, 318, 12402]])
- * this means the next value we want is 341
- */
- /*while(last_val!=341) {
+ last_val = predict_next_token(27017);
- train_mask = last_val ^ 341;
+ /* Compare values
+ * The phrase "Taxation is theft"
+ * tokenized with the GPT2 tokenizer
+ * is tensor([[27017, 341, 318, 12402]])
+ * this means the next value we want is 341
+ */
+ while(last_val!=341) {
- print("Run training single shot: ");
- citoa(train_mask, numstr, 10);
- print(numstr);
- ack();
-
- //set_dws(dw);
- set_dws(dw, false);
- dw=dw/100;
- dw=99*dw;
- mask_back_propgatation(train_mask);
-
- citoa(27017, numstr, 10);
- print("Input value: ");
- print(numstr);
- ack();
-
- last_val = predict_next_token(27017);
+ train_mask = last_val ^ 341;
- citoa(last_val, numstr, 10);
- print("Output value: ");
- print(numstr);
- ack();
- }*/
+ //set_dws(dw);
+ set_dws(dw, false);
+ dw=dw/100;
+ dw=99*dw;
+ mask_back_propgatation(train_mask);
- //print("EXIT");
- //ack();
+ last_val = predict_next_token(27017);
+ }
-#endif
+}
diff --git a/src/py/tty3.py b/src/py/tty3.py
new file mode 100644
index 0000000..7ff1660
--- /dev/null
+++ b/src/py/tty3.py
@@ -0,0 +1,33 @@
+import json
+
+from fac_tools import run_command
+from fac_tools import get_fac_wrapper
+from fac_tools import load_weights_and_biases
+
+from transformers import AutoTokenizer
+
+server = get_fac_wrapper("telnet")
+
+print("Reading in JSON")
+with open("test_files/weights_and_biases.json", "r") as f:
+ weights_and_biases = json.loads(f.read())
+ f.close()
+
+#load_weights_and_biases(server, weights_and_biases)
+
+tokenizer = AutoTokenizer.from_pretrained("gpt2")
+prompt = "Taxation is theft"
+input_ids = tokenizer(prompt, return_tensors="pt").input_ids
+tokens = input_ids[0].numpy()
+run_command(server,"TRAIN")
+run_command(server,"TOKENS")
+for tok in tokens:
+ run_command(server,str(tok))
+run_command(server,"DONE")
+
+run_command(server,"TRAIN")
+run_command(server,"SHOOT")
+
+run_command(server,"TERMINATE")
+
+server.close()
--
GitLab