Upload nand.py
Browse files
nand.py
ADDED
@@ -0,0 +1,349 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from huggingface_hub import InferenceClient
|
2 |
+
from langchain_community.embeddings.sentence_transformer import SentenceTransformerEmbeddings
|
3 |
+
from langchain_community.vectorstores import Chroma
|
4 |
+
from transformers import pipeline
|
5 |
+
from sentence_transformers.cross_encoder import CrossEncoder
|
6 |
+
import re
|
7 |
+
|
8 |
+
def setupDB(domain, hasLLM):
|
9 |
+
history = []
|
10 |
+
history.append("")
|
11 |
+
history.append("")
|
12 |
+
crossmodel = CrossEncoder("cross-encoder/stsb-distilroberta-base")
|
13 |
+
models,allState = nandState()
|
14 |
+
support_db = nandGetChroma(domain)
|
15 |
+
|
16 |
+
insts_db = nandGetChroma("insts")
|
17 |
+
|
18 |
+
|
19 |
+
pdf_dbs = []
|
20 |
+
if domain == 'en':
|
21 |
+
pdfs = [] #"pdf_0em", "pdf_1em", "pdf_2em", "pdf_3em","pdf_4em"]
|
22 |
+
for onepdf in pdfs:
|
23 |
+
pdfdb = nandGetChroma(onepdf)
|
24 |
+
pdf_dbs.append(pdfdb)
|
25 |
+
para = {}
|
26 |
+
para['history'] = history
|
27 |
+
para['disnum'] = 10
|
28 |
+
para['domain'] = domain
|
29 |
+
para['crossmodel'] = crossmodel
|
30 |
+
para['insts_db'] = insts_db
|
31 |
+
para['support_db'] = support_db
|
32 |
+
para['pdf_dbs'] = pdf_dbs
|
33 |
+
para['hasLLM'] = hasLLM
|
34 |
+
return para
|
35 |
+
def remapScore(domain, inscore):
|
36 |
+
if domain == 'ch':
|
37 |
+
xin = 1 - inscore
|
38 |
+
a = -0.2
|
39 |
+
b = 1.2
|
40 |
+
y = a * xin * xin + b * xin
|
41 |
+
return int(y * 100)
|
42 |
+
else:
|
43 |
+
xin = 1 - inscore
|
44 |
+
a = -1.2
|
45 |
+
b = 2.2
|
46 |
+
y = a * xin * xin + b * xin
|
47 |
+
return int(y * 100)
|
48 |
+
|
49 |
+
def process_query(iniquery, para):
|
50 |
+
query = re.sub("<br>", "", iniquery)
|
51 |
+
ch2en, query = toEn(query)
|
52 |
+
if ch2en:
|
53 |
+
print(f"Received from connected users : {query}")
|
54 |
+
else:
|
55 |
+
print(f"Received from connected users : {query}", end='')
|
56 |
+
disnum = para['disnum']
|
57 |
+
domain = para['domain']
|
58 |
+
history = para['history']
|
59 |
+
crossmodel = para['crossmodel']
|
60 |
+
insts_db = para['insts_db']
|
61 |
+
support_db = para['support_db']
|
62 |
+
pdf_dbs = para['pdf_dbs']
|
63 |
+
hasLLM = para['hasLLM']
|
64 |
+
ret = ""
|
65 |
+
|
66 |
+
needScriptScores = crossmodel.predict([["write a perl ECO script", query]])
|
67 |
+
print(f"THE QUERY SCORE for creating eco script: score={needScriptScores[0]}")
|
68 |
+
allapis = []
|
69 |
+
threshold = 0.45
|
70 |
+
if needScriptScores[0] > threshold:
|
71 |
+
print(f"THE QUERY REQUIRES CREATING AN ECO SCRIPT score={needScriptScores[0]} > {threshold}")
|
72 |
+
retinsts = insts_db.similarity_search_with_score(query, k=10)
|
73 |
+
accu = 0
|
74 |
+
for inst in retinsts:
|
75 |
+
instdoc = inst[0]
|
76 |
+
instscore = inst[1]
|
77 |
+
instname = instdoc.metadata['source']
|
78 |
+
otherfile = re.sub("^insts", "src_en", instname)
|
79 |
+
otherfile = re.sub("\.\d+", "", otherfile)
|
80 |
+
if not otherfile in allapis:
|
81 |
+
allapis.append(otherfile)
|
82 |
+
apisize = os.path.getsize(otherfile)
|
83 |
+
accu += apisize
|
84 |
+
print(f"INST: {instname} SCORE: {instscore} API-size: {apisize} Accu: {accu}")
|
85 |
+
|
86 |
+
results = []
|
87 |
+
docs = support_db.similarity_search_with_score(query, k=8)
|
88 |
+
for doc in docs:
|
89 |
+
results.append([doc[0], doc[1]])
|
90 |
+
for onepdfdb in pdf_dbs:
|
91 |
+
pdocs = onepdfdb.similarity_search_with_score(query, k=8)
|
92 |
+
for doc in pdocs:
|
93 |
+
results.append([doc[0], doc[1]+0.2])
|
94 |
+
results.sort(key=lambda x: x[1])
|
95 |
+
docnum = len(results)
|
96 |
+
index = 1
|
97 |
+
for ii in range(docnum):
|
98 |
+
doc = results[ii][0]
|
99 |
+
source = doc.metadata['source']
|
100 |
+
path = source #source.replace("\\", "/")
|
101 |
+
#print(f"path={path}")
|
102 |
+
if path in allapis:
|
103 |
+
print(f"dont use path={path}, it's in instruction list")
|
104 |
+
continue
|
105 |
+
prefix = "Help:"
|
106 |
+
if re.search("api\.", source):
|
107 |
+
prefix = "API:"
|
108 |
+
elif re.search("man\.", source):
|
109 |
+
prefix = "Manual:"
|
110 |
+
elif re.search("\.pdf$", source):
|
111 |
+
prefix = "PDF:";
|
112 |
+
score = remapScore(domain, results[ii][1])
|
113 |
+
retcont = doc.page_content
|
114 |
+
if re.search("\.pdf$", source):
|
115 |
+
page = doc.metadata['page'] + 1
|
116 |
+
subpage = doc.metadata['subpage']
|
117 |
+
retcont += f"\n<a target='_blank' href='/AI/{path}#page={page}'>PDF{page} {subpage}</a>\n"
|
118 |
+
ret += f"Return {index} ({score}) {prefix} {retcont}\n"
|
119 |
+
if len(ret) > 6000:
|
120 |
+
break
|
121 |
+
index += 1
|
122 |
+
if index > disnum:
|
123 |
+
break
|
124 |
+
if hasLLM:
|
125 |
+
context = "Context information is below\n---------------------\n"
|
126 |
+
if len(allapis):
|
127 |
+
context += scriptExamples()
|
128 |
+
for oneapi in allapis:
|
129 |
+
cont = GetContent(oneapi)
|
130 |
+
cont = re.sub("</h3>", " API Detail:", cont)
|
131 |
+
cont = re.sub('<.*?>', '', cont)
|
132 |
+
cont = re.sub('Examples:.*', '', cont, flags=re.DOTALL)
|
133 |
+
context += cont
|
134 |
+
context += ret
|
135 |
+
prompt = f"{context}\n"
|
136 |
+
prompt += "------------------------------------------\n"
|
137 |
+
if len(allapis):
|
138 |
+
prompt += "Given the context information and not prior knowledge, creat a Perl ECO script by following the format and sequence in the script examples provided above.\n"
|
139 |
+
#prompt += "1. Following the format in the script examples provided above.\n"
|
140 |
+
#prompt += "2. Following the API sequence in the script examples above, for instance, APIs get_spare_cells and map_spare_cells should be after fix_design.\n"
|
141 |
+
else:
|
142 |
+
prompt += "Given the context information and not prior knowledge, answer the query.\n"
|
143 |
+
prompt += f"Query: {query}\n"
|
144 |
+
|
145 |
+
llmout = llmGenerate(prompt)
|
146 |
+
history[0] = query
|
147 |
+
history[1] = llmout
|
148 |
+
#return llmout
|
149 |
+
outlen = len(llmout)
|
150 |
+
prolen = len(prompt)
|
151 |
+
print(f"Prompt len: {prolen} LLMOUT len: {outlen}")
|
152 |
+
allret = "LLM_OUTPUT_START:"+llmout+"\nEND OF LLM OUTPUT\n"+prompt
|
153 |
+
return allret
|
154 |
+
return ret
|
155 |
+
|
156 |
+
def toEn(intxt):
|
157 |
+
pattern = re.compile(r'[\u4e00-\u9fff]+')
|
158 |
+
if pattern.search(intxt):
|
159 |
+
translator = pipeline(task="translation", model="Helsinki-NLP/opus-mt-zh-en")
|
160 |
+
ini_text = translator(intxt, max_length=500)[0]['translation_text']
|
161 |
+
out_text = re.sub("ECO foot", "ECO Script", ini_text)
|
162 |
+
out_text = re.sub("web-based", "netlist", out_text)
|
163 |
+
out_text = re.sub(r"\bweb\b", "netlist", out_text)
|
164 |
+
out_text = re.sub(r"\bwebsheet\b", "netlist", out_text)
|
165 |
+
out_text = re.sub(r"\bweblists?\b", "netlist", out_text)
|
166 |
+
print(f"AFTER RESULT: {out_text}")
|
167 |
+
return 1, out_text
|
168 |
+
return 0, intxt
|
169 |
+
|
170 |
+
|
171 |
+
|
172 |
+
def nandGetChroma(domain):
|
173 |
+
models,allState = nandState()
|
174 |
+
chdb = allState[domain]['chroma']
|
175 |
+
print(f"domain: {domain} has chroma dir {chdb}")
|
176 |
+
model_ind = allState[domain]['model']
|
177 |
+
model_name = models[model_ind]
|
178 |
+
embedding_function = SentenceTransformerEmbeddings(model_name=model_name)
|
179 |
+
chroma_db = Chroma(persist_directory=chdb, embedding_function=embedding_function)
|
180 |
+
return chroma_db
|
181 |
+
def nandState():
|
182 |
+
models = {'em': "all-MiniLM-L6-v2",
|
183 |
+
'en': "all-mpnet-base-v2",
|
184 |
+
'ch': "shibing624/text2vec-base-chinese-sentence"}
|
185 |
+
# chunk is to cut the big PDF page to smaller, 1000byte chunks, and chinese page into smaller chunks
|
186 |
+
allState = {'insts':{'cstate':{},'pstate':{},'dir':'insts','json':'filestatus.insts.json','chroma':'chroma_db_insts','model':'en','chunk':0},
|
187 |
+
'en':{'cstate':{},'pstate':{},'dir':'src_en','json':'filestatus.english.json','chroma':'chroma_db_en','model':'en','chunk':0},
|
188 |
+
'ch':{'cstate':{},'pstate':{},'dir':'src_ch','json':'filestatus.chinese.json','chroma':'chroma_db_ch','model':'ch','chunk':1}
|
189 |
+
}
|
190 |
+
|
191 |
+
for ind in range(12):
|
192 |
+
name = f"pdf_{ind}em"
|
193 |
+
allState[name] = {'cstate':{},'pstate':{},'dir':f"pdf_sub{ind}",'json':f"filestatus.{name}.json",'chroma':f"chroma_db_{name}",'model':'em','chunk':1}
|
194 |
+
return models, allState
|
195 |
+
def formatPrompt(message, history):
|
196 |
+
if history[0]:
|
197 |
+
prompt = "Create a new query based on previous query/answer paire and current query:\n"
|
198 |
+
prompt += f"Previous query: {history[0]}"
|
199 |
+
prompt += f"Previous answer: {histroy[1]}"
|
200 |
+
prompt += f"Current query: {message}"
|
201 |
+
prompt += "New query:"
|
202 |
+
return prompt
|
203 |
+
return message
|
204 |
+
|
205 |
+
def llmNewQuery(prompt, history):
|
206 |
+
newpend = formatPrompt(prompt, history)
|
207 |
+
newquery = llmGenerate(newpend)
|
208 |
+
return newquery
|
209 |
+
|
210 |
+
def llmGenerate(prompt, temperature=0.001, max_new_tokens=2048, top_p=0.95, repetition_penalty=1.0):
|
211 |
+
#temperature = float(temperature)
|
212 |
+
#if temperature < 1e-2:
|
213 |
+
# temperature = 1e-2
|
214 |
+
top_p = float(top_p)
|
215 |
+
|
216 |
+
generate_kwargs = dict(
|
217 |
+
temperature=temperature,
|
218 |
+
max_new_tokens=max_new_tokens,
|
219 |
+
top_p=top_p,
|
220 |
+
repetition_penalty=repetition_penalty,
|
221 |
+
do_sample=True,
|
222 |
+
seed=42,
|
223 |
+
)
|
224 |
+
llmclient = InferenceClient("mistralai/Mistral-7B-Instruct-v0.2")
|
225 |
+
|
226 |
+
stream = llmclient.text_generation(prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
227 |
+
output = ""
|
228 |
+
|
229 |
+
for response in stream:
|
230 |
+
output += response.token.text
|
231 |
+
#yield output
|
232 |
+
return output
|
233 |
+
|
234 |
+
|
235 |
+
def thoseRemove():
|
236 |
+
those = ["www.synopsys.com", "sy ?nopsys", "cadence", "mentor", "solvnetplus", "solvnet"]
|
237 |
+
return those
|
238 |
+
|
239 |
+
def GetContent(file):
|
240 |
+
fcont = ""
|
241 |
+
with open(file) as f:
|
242 |
+
fcont = f.read()
|
243 |
+
return fcont
|
244 |
+
|
245 |
+
def scriptExamples():
|
246 |
+
exp = """
|
247 |
+
#The first ECO scipt example for manual ECO:
|
248 |
+
use strict;
|
249 |
+
setup_eco("eco_example");
|
250 |
+
read_library("tsmc.5nm.lib");
|
251 |
+
read_design("-imp", "implementation.gv");
|
252 |
+
set_top("topmod");
|
253 |
+
change_pin("u_abc/state_reg_0_/D", "INVX1", "", "-");
|
254 |
+
change_pin("u_abc/state_reg_1_/D", "INVX1", "", "-");
|
255 |
+
change_pin("u_abc/state_reg_2_/D", "INVX1", "", "-");
|
256 |
+
report_eco(); # ECO report
|
257 |
+
check_design();
|
258 |
+
write_verilog("eco_verilog.v");# Write out ECO result in Verilog
|
259 |
+
#End of the manual ECO script example
|
260 |
+
|
261 |
+
#The second ECO script example for automatic ECO:
|
262 |
+
use strict;
|
263 |
+
setup_eco("eco_example");# Setup ECO name
|
264 |
+
read_library("tsmc.5nm.lib");# Read in standard library
|
265 |
+
# SVF files are optional, best to be used when the design involves multibit flops
|
266 |
+
#read_svf("-ref", "reference.svf.txt");
|
267 |
+
#read_svf("-imp", "implementation.svf.txt");
|
268 |
+
read_design("-ref", "reference.gv");
|
269 |
+
read_design("-imp", "implementation.gv");
|
270 |
+
set_top("topmod");# Set the top module
|
271 |
+
# Preserve DFT Test Logic
|
272 |
+
set_ignore_output("scan_out*");
|
273 |
+
set_pin_constant("scan_enable", 0);
|
274 |
+
set_pin_constant("scan_mode", 0);
|
275 |
+
fix_design();
|
276 |
+
report_eco(); # ECO report
|
277 |
+
check_design();
|
278 |
+
write_verilog("eco_verilog.v");# Write out ECO result in Verilog
|
279 |
+
run_lec(); # Run GOF LEC to generate Formality help files
|
280 |
+
#End of automatic ECO script example
|
281 |
+
|
282 |
+
|
283 |
+
#The third ECO script example is for automatic metal only ECO:
|
284 |
+
use strict;
|
285 |
+
setup_eco("eco_example");# Setup ECO name
|
286 |
+
read_library("tsmc.5nm.lib");# Read in standard library
|
287 |
+
# SVF files are optional, best to be used when the design involves multibit flops
|
288 |
+
#read_svf("-ref", "reference.svf.txt");
|
289 |
+
#read_svf("-imp", "implementation.svf.txt");
|
290 |
+
read_design("-ref", "reference.gv");# Read in Reference Netlist
|
291 |
+
read_design("-imp", "implementation.gv");
|
292 |
+
set_top("topmod");# Set the top module
|
293 |
+
set_ignore_output("scan_out*");
|
294 |
+
set_pin_constant("scan_enable", 0);
|
295 |
+
set_pin_constant("scan_mode", 0);
|
296 |
+
read_lef("tsmc.lef"); # Read LEF
|
297 |
+
read_def("topmod.def"); # Read Design Exchange Format file
|
298 |
+
fix_design(); # Must run before get_spare_cells and map_spare_cells
|
299 |
+
get_spare_cells("*/*_SPARE*");
|
300 |
+
map_spare_cells();
|
301 |
+
report_eco(); # ECO report
|
302 |
+
check_design();# Check if the ECO causes any issue, like floating
|
303 |
+
write_verilog("eco_verilog.v");# Write out ECO result in Verilog
|
304 |
+
write_perl("eco_result.pl");# Write out result in Perl script
|
305 |
+
run_lec(); # Run GOF LEC to generate Formality help files
|
306 |
+
#End of automatic ECO script example
|
307 |
+
|
308 |
+
#The four ECO script example is the same as the third ECO script, except fix_design
|
309 |
+
# list_file option to load in the ECO points list file converted from RTL-to-RTL LEC result
|
310 |
+
fix_design("-list_file", "the_eco_points.txt");
|
311 |
+
|
312 |
+
#The 5th ECO script example is the same as the 3rd ECO script, except fix_design
|
313 |
+
# Enable flatten mode ECO. The default mode is hierarchical. The flatten mode is for small fix but the changes go across
|
314 |
+
# module boundaries
|
315 |
+
fix_design("-flatten");
|
316 |
+
|
317 |
+
#The 6th ECO script is similar to the third ECO script, but it dumps formality help file after LEC
|
318 |
+
run_lec(); # Run GOF LEC to generate Formality help files
|
319 |
+
write_compare_points("compare_points.report");
|
320 |
+
write_formality_help_files("fm_dir/formality_help"); # formality_help files are generated in fm_dir folder
|
321 |
+
|
322 |
+
#The 7th ECO script is similar to the third ECO script, but it uses gate array spare cells
|
323 |
+
fix_design(); # Must run before get_spare_cells and map_spare_cells
|
324 |
+
# Enable Gate Array Spare Cells Metal Only ECO Flow, map_spare_cells will map to Gate Array Cells only
|
325 |
+
get_spare_cells("-gate_array", "G*", "-gate_array_filler", "GFILL*|GDCAP*");
|
326 |
+
map_spare_cells();
|
327 |
+
|
328 |
+
#The 8th ECO script is similar to the third ECO script, but it uses only deleted gates or freed up gates in ECO as spare cells
|
329 |
+
fix_design(); # Must run before get_spare_cells and map_spare_cells
|
330 |
+
get_spare_cells("-addfreed");
|
331 |
+
map_spare_cells();
|
332 |
+
|
333 |
+
#The 9th ECO script is manual ECO, find all memory hierarchically and tie the pin TEST_SHIFT of memory to net "TEST_EN"
|
334 |
+
use strict;
|
335 |
+
setup_eco("eco_example");
|
336 |
+
read_library("tsmc.3nm.lib");
|
337 |
+
read_design("-imp", "from_backend.gv");
|
338 |
+
set_top("topmod");
|
339 |
+
# Get all memories hierarchically, instance naming, "U_HMEM*"
|
340 |
+
my @mems = get_cells("-hier", "U_HMEM*");
|
341 |
+
foreach my $mem (@mems){
|
342 |
+
change_pin("$mem/TEST_SHIFT", "TEST_EN");
|
343 |
+
}
|
344 |
+
report_eco(); # ECO report
|
345 |
+
check_design();
|
346 |
+
write_verilog("mem_eco.v");
|
347 |
+
|
348 |
+
"""
|
349 |
+
return exp
|