kmfoda commited on
Commit
0f87068
·
1 Parent(s): 2d26195

Update results

Browse files
Files changed (3) hide show
  1. app.py +2 -1
  2. evaluate.py +4 -3
  3. results.json +723 -0
app.py CHANGED
@@ -1,5 +1,6 @@
1
- import gradio as gr
2
  import json
 
 
3
  import pandas as pd
4
 
5
  with open('results.json', 'r') as file:
 
 
1
  import json
2
+
3
+ import gradio as gr
4
  import pandas as pd
5
 
6
  with open('results.json', 'r') as file:
evaluate.py CHANGED
@@ -1,9 +1,10 @@
 
 
 
1
  import torch
2
- from transformers import AutoModelForCausalLM, AutoTokenizer
3
  from distributed_training.data.dataset import DataLoader
4
- import random
5
  from huggingface_hub import list_repo_refs
6
- import json
7
 
8
  device = "cuda"
9
  test_indices_length = 10
 
1
+ import json
2
+ import random
3
+
4
  import torch
 
5
  from distributed_training.data.dataset import DataLoader
 
6
  from huggingface_hub import list_repo_refs
7
+ from transformers import AutoModelForCausalLM, AutoTokenizer
8
 
9
  device = "cuda"
10
  test_indices_length = 10
results.json CHANGED
@@ -1895,6 +1895,729 @@
1895
  ],
1896
  "631": [
1897
  6.486404657363892
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1898
  ]
1899
  },
1900
  "distributed/optimized-gpt2-250m-v0.1.1": {
 
1895
  ],
1896
  "631": [
1897
  6.486404657363892
1898
+ ],
1899
+ "632": [
1900
+ 6.643767833709717
1901
+ ],
1902
+ "633": [
1903
+ 6.1109654903411865
1904
+ ],
1905
+ "634": [
1906
+ 6.537951628367106
1907
+ ],
1908
+ "635": [
1909
+ 6.427129173278809
1910
+ ],
1911
+ "636": [
1912
+ 6.15797483921051
1913
+ ],
1914
+ "637": [
1915
+ 5.996870803833008
1916
+ ],
1917
+ "638": [
1918
+ 6.446497281392415
1919
+ ],
1920
+ "639": [
1921
+ 6.442559242248535
1922
+ ],
1923
+ "640": [
1924
+ 6.861009216308593
1925
+ ],
1926
+ "641": [
1927
+ 6.906637787818909
1928
+ ],
1929
+ "642": [
1930
+ 6.548080046971639
1931
+ ],
1932
+ "643": [
1933
+ 6.7596203804016115
1934
+ ],
1935
+ "644": [
1936
+ 6.254246711730957
1937
+ ],
1938
+ "645": [
1939
+ 6.172244707743327
1940
+ ],
1941
+ "646": [
1942
+ 6.69602358341217
1943
+ ],
1944
+ "647": [
1945
+ 6.616944630940755
1946
+ ],
1947
+ "648": [
1948
+ 6.462185025215149
1949
+ ],
1950
+ "649": [
1951
+ 6.446821975708008
1952
+ ],
1953
+ "650": [
1954
+ 6.728310585021973
1955
+ ],
1956
+ "651": [
1957
+ 6.57671730858939
1958
+ ],
1959
+ "652": [
1960
+ 6.995179653167725
1961
+ ],
1962
+ "653": [
1963
+ 6.282734680175781
1964
+ ],
1965
+ "654": [
1966
+ 6.708661317825317
1967
+ ],
1968
+ "655": [
1969
+ 6.749908447265625
1970
+ ],
1971
+ "656": [
1972
+ 6.361384868621826
1973
+ ],
1974
+ "657": [
1975
+ 5.6892744700113935
1976
+ ],
1977
+ "658": [
1978
+ 6.620821339743478
1979
+ ],
1980
+ "659": [
1981
+ 6.714852929115295
1982
+ ],
1983
+ "660": [
1984
+ 6.357420921325684
1985
+ ],
1986
+ "661": [
1987
+ 6.2726335525512695
1988
+ ],
1989
+ "662": [
1990
+ 6.467626285552979
1991
+ ],
1992
+ "663": [
1993
+ 6.681742310523987
1994
+ ],
1995
+ "664": [
1996
+ 6.497158368428548
1997
+ ],
1998
+ "665": [
1999
+ 6.731684303283691
2000
+ ],
2001
+ "666": [
2002
+ 6.617751916249593
2003
+ ],
2004
+ "667": [
2005
+ 6.470035711924235
2006
+ ],
2007
+ "668": [
2008
+ 6.931012153625488
2009
+ ],
2010
+ "669": [
2011
+ 6.415357249123709
2012
+ ],
2013
+ "670": [
2014
+ 6.616030851999919
2015
+ ],
2016
+ "671": [
2017
+ 6.5195033073425295
2018
+ ],
2019
+ "672": [
2020
+ 6.511032899220784
2021
+ ],
2022
+ "673": [
2023
+ 6.45783535639445
2024
+ ],
2025
+ "674": [
2026
+ 6.389084339141846
2027
+ ],
2028
+ "675": [
2029
+ 6.980731725692749
2030
+ ],
2031
+ "676": [
2032
+ 6.267537832260132
2033
+ ],
2034
+ "677": [
2035
+ 6.549607038497925
2036
+ ],
2037
+ "678": [
2038
+ 6.524194145202637
2039
+ ],
2040
+ "679": [
2041
+ 6.827434380849202
2042
+ ],
2043
+ "680": [
2044
+ 6.130102157592773
2045
+ ],
2046
+ "681": [
2047
+ 6.263987143834432
2048
+ ],
2049
+ "682": [
2050
+ 6.9984735488891605
2051
+ ],
2052
+ "683": [
2053
+ 6.189664522806804
2054
+ ],
2055
+ "684": [
2056
+ 6.498591581980388
2057
+ ],
2058
+ "685": [
2059
+ 6.275699751717704
2060
+ ],
2061
+ "686": [
2062
+ 6.250300765037537
2063
+ ],
2064
+ "687": [
2065
+ 6.450555801391602
2066
+ ],
2067
+ "688": [
2068
+ 6.374192555745442
2069
+ ],
2070
+ "689": [
2071
+ 6.7119139432907104
2072
+ ],
2073
+ "690": [
2074
+ 6.6327869415283205
2075
+ ],
2076
+ "691": [
2077
+ 6.440953922271729
2078
+ ],
2079
+ "692": [
2080
+ 6.211158752441406
2081
+ ],
2082
+ "693": [
2083
+ 6.941821813583374
2084
+ ],
2085
+ "694": [
2086
+ 6.48214282989502
2087
+ ],
2088
+ "695": [
2089
+ 6.502297759056091
2090
+ ],
2091
+ "696": [
2092
+ 6.2446370124816895
2093
+ ],
2094
+ "697": [
2095
+ 6.4006937344868975
2096
+ ],
2097
+ "698": [
2098
+ 6.494188189506531
2099
+ ],
2100
+ "699": [
2101
+ 6.883090019226074
2102
+ ],
2103
+ "700": [
2104
+ 6.237102150917053
2105
+ ],
2106
+ "701": [
2107
+ 6.412813901901245
2108
+ ],
2109
+ "702": [
2110
+ 6.395236094792684
2111
+ ],
2112
+ "703": [
2113
+ 6.461537917455037
2114
+ ],
2115
+ "704": [
2116
+ 6.824586629867554
2117
+ ],
2118
+ "705": [
2119
+ 6.02636182308197
2120
+ ],
2121
+ "706": [
2122
+ 6.476898193359375
2123
+ ],
2124
+ "707": [
2125
+ 6.428057551383972
2126
+ ],
2127
+ "708": [
2128
+ 6.534157037734985
2129
+ ],
2130
+ "709": [
2131
+ 6.673372387886047
2132
+ ],
2133
+ "710": [
2134
+ 6.520640850067139
2135
+ ],
2136
+ "711": [
2137
+ 6.776037693023682
2138
+ ],
2139
+ "712": [
2140
+ 6.543991923332214
2141
+ ],
2142
+ "713": [
2143
+ 6.379572868347168
2144
+ ],
2145
+ "714": [
2146
+ 5.9953097105026245
2147
+ ],
2148
+ "715": [
2149
+ 6.289546966552734
2150
+ ],
2151
+ "716": [
2152
+ 6.783546447753906
2153
+ ],
2154
+ "717": [
2155
+ 6.394249677658081
2156
+ ],
2157
+ "718": [
2158
+ 6.733166933059692
2159
+ ],
2160
+ "719": [
2161
+ 6.117873589197795
2162
+ ],
2163
+ "720": [
2164
+ 6.275993537902832
2165
+ ],
2166
+ "721": [
2167
+ 6.235243439674377
2168
+ ],
2169
+ "722": [
2170
+ 6.411739826202393
2171
+ ],
2172
+ "723": [
2173
+ 6.584866523742676
2174
+ ],
2175
+ "724": [
2176
+ 6.358508944511414
2177
+ ],
2178
+ "725": [
2179
+ 6.375350634256999
2180
+ ],
2181
+ "726": [
2182
+ 6.483253002166748
2183
+ ],
2184
+ "727": [
2185
+ 6.6668111483256025
2186
+ ],
2187
+ "728": [
2188
+ 6.593089962005616
2189
+ ],
2190
+ "729": [
2191
+ 6.094300190607707
2192
+ ],
2193
+ "730": [
2194
+ 5.817228889465332
2195
+ ],
2196
+ "731": [
2197
+ 6.6721724510192875
2198
+ ],
2199
+ "732": [
2200
+ 6.248234033584595
2201
+ ],
2202
+ "733": [
2203
+ 6.197977352142334
2204
+ ],
2205
+ "734": [
2206
+ 6.409354591369629
2207
+ ],
2208
+ "735": [
2209
+ 6.680954813957214
2210
+ ],
2211
+ "736": [
2212
+ 6.469669342041016
2213
+ ],
2214
+ "737": [
2215
+ 6.020197629928589
2216
+ ],
2217
+ "738": [
2218
+ 6.1867780685424805
2219
+ ],
2220
+ "739": [
2221
+ 6.147508502006531
2222
+ ],
2223
+ "740": [
2224
+ 6.529857397079468
2225
+ ],
2226
+ "741": [
2227
+ 6.497222900390625
2228
+ ],
2229
+ "742": [
2230
+ 5.965197324752808
2231
+ ],
2232
+ "743": [
2233
+ 6.157637715339661
2234
+ ],
2235
+ "744": [
2236
+ 5.941549682617188
2237
+ ],
2238
+ "745": [
2239
+ 6.416422128677368
2240
+ ],
2241
+ "746": [
2242
+ 6.185433626174927
2243
+ ],
2244
+ "747": [
2245
+ 6.388829867045085
2246
+ ],
2247
+ "748": [
2248
+ 6.51822566986084
2249
+ ],
2250
+ "749": [
2251
+ 6.221700191497803
2252
+ ],
2253
+ "750": [
2254
+ 6.086953067779541
2255
+ ],
2256
+ "751": [
2257
+ 6.252593517303467
2258
+ ],
2259
+ "752": [
2260
+ 6.464523029327393
2261
+ ],
2262
+ "753": [
2263
+ 6.2163978815078735
2264
+ ],
2265
+ "754": [
2266
+ 6.640975634256999
2267
+ ],
2268
+ "755": [
2269
+ 6.48329496383667
2270
+ ],
2271
+ "756": [
2272
+ 6.418565630912781
2273
+ ],
2274
+ "757": [
2275
+ 5.7117462158203125
2276
+ ],
2277
+ "758": [
2278
+ 6.1879366636276245
2279
+ ],
2280
+ "759": [
2281
+ 6.296315312385559
2282
+ ],
2283
+ "760": [
2284
+ 6.109013199806213
2285
+ ],
2286
+ "761": [
2287
+ 5.8764106432596845
2288
+ ],
2289
+ "762": [
2290
+ 6.327104727427165
2291
+ ],
2292
+ "763": [
2293
+ 5.999971628189087
2294
+ ],
2295
+ "764": [
2296
+ 6.284005355834961
2297
+ ],
2298
+ "765": [
2299
+ 6.335148016611735
2300
+ ],
2301
+ "766": [
2302
+ 5.732976754506429
2303
+ ],
2304
+ "767": [
2305
+ 6.240153153737386
2306
+ ],
2307
+ "768": [
2308
+ 6.102706074714661
2309
+ ],
2310
+ "769": [
2311
+ 6.126063585281372
2312
+ ],
2313
+ "770": [
2314
+ 6.413216193517049
2315
+ ],
2316
+ "771": [
2317
+ 6.535668849945068
2318
+ ],
2319
+ "772": [
2320
+ 6.662901878356934
2321
+ ],
2322
+ "773": [
2323
+ 5.962766249974568
2324
+ ],
2325
+ "774": [
2326
+ 6.299613761901855
2327
+ ],
2328
+ "775": [
2329
+ 6.219697093963623
2330
+ ],
2331
+ "776": [
2332
+ 6.310187657674153
2333
+ ],
2334
+ "777": [
2335
+ 6.4950854778289795
2336
+ ],
2337
+ "778": [
2338
+ 6.248437801996867
2339
+ ],
2340
+ "779": [
2341
+ 6.183503150939941
2342
+ ],
2343
+ "780": [
2344
+ 6.268959124883016
2345
+ ],
2346
+ "781": [
2347
+ 6.388804197311401
2348
+ ],
2349
+ "782": [
2350
+ 5.9852478981018065
2351
+ ],
2352
+ "783": [
2353
+ 6.184964299201965
2354
+ ],
2355
+ "784": [
2356
+ 6.189103317260742
2357
+ ],
2358
+ "785": [
2359
+ 6.4577540556589765
2360
+ ],
2361
+ "786": [
2362
+ 6.466018358866374
2363
+ ],
2364
+ "787": [
2365
+ 6.145304361979167
2366
+ ],
2367
+ "788": [
2368
+ 6.428526242574056
2369
+ ],
2370
+ "789": [
2371
+ 6.3111328125
2372
+ ],
2373
+ "790": [
2374
+ 6.128303289413452
2375
+ ],
2376
+ "791": [
2377
+ 6.2964760065078735
2378
+ ],
2379
+ "792": [
2380
+ 6.701506932576497
2381
+ ],
2382
+ "793": [
2383
+ 6.1448437690734865
2384
+ ],
2385
+ "794": [
2386
+ 5.943706671396892
2387
+ ],
2388
+ "795": [
2389
+ 6.1738550662994385
2390
+ ],
2391
+ "796": [
2392
+ 6.041583061218262
2393
+ ],
2394
+ "797": [
2395
+ 6.03037961324056
2396
+ ],
2397
+ "798": [
2398
+ 6.437365770339966
2399
+ ],
2400
+ "799": [
2401
+ 5.87385356426239
2402
+ ],
2403
+ "800": [
2404
+ 6.0634894371032715
2405
+ ],
2406
+ "801": [
2407
+ 6.383962631225586
2408
+ ],
2409
+ "802": [
2410
+ 6.762539267539978
2411
+ ],
2412
+ "803": [
2413
+ 6.493936896324158
2414
+ ],
2415
+ "804": [
2416
+ 6.575284481048584
2417
+ ],
2418
+ "805": [
2419
+ 6.2409255027771
2420
+ ],
2421
+ "806": [
2422
+ 6.42514123916626
2423
+ ],
2424
+ "807": [
2425
+ 6.345109065373738
2426
+ ],
2427
+ "808": [
2428
+ 6.135688781738281
2429
+ ],
2430
+ "809": [
2431
+ 6.439762592315674
2432
+ ],
2433
+ "810": [
2434
+ 6.485467195510864
2435
+ ],
2436
+ "811": [
2437
+ 6.508481979370117
2438
+ ],
2439
+ "812": [
2440
+ 6.524493312835693
2441
+ ],
2442
+ "813": [
2443
+ 5.99509612719218
2444
+ ],
2445
+ "814": [
2446
+ 5.73370361328125
2447
+ ],
2448
+ "815": [
2449
+ 6.451435565948486
2450
+ ],
2451
+ "816": [
2452
+ 6.277082920074463
2453
+ ],
2454
+ "817": [
2455
+ 6.068041960398356
2456
+ ],
2457
+ "818": [
2458
+ 6.102489153544108
2459
+ ],
2460
+ "819": [
2461
+ 6.086841011047364
2462
+ ],
2463
+ "820": [
2464
+ 6.422192215919495
2465
+ ],
2466
+ "821": [
2467
+ 6.113222241401672
2468
+ ],
2469
+ "822": [
2470
+ 6.005942344665527
2471
+ ],
2472
+ "823": [
2473
+ 6.628415775299072
2474
+ ],
2475
+ "824": [
2476
+ 6.376164078712463
2477
+ ],
2478
+ "825": [
2479
+ 6.349189281463623
2480
+ ],
2481
+ "826": [
2482
+ 6.0373135805130005
2483
+ ],
2484
+ "827": [
2485
+ 5.850473523139954
2486
+ ],
2487
+ "828": [
2488
+ 6.014297246932983
2489
+ ],
2490
+ "829": [
2491
+ 6.12701530456543
2492
+ ],
2493
+ "830": [
2494
+ 6.539065837860107
2495
+ ],
2496
+ "831": [
2497
+ 6.135456562042236
2498
+ ],
2499
+ "832": [
2500
+ 6.186184406280518
2501
+ ],
2502
+ "833": [
2503
+ 6.082632780075073
2504
+ ],
2505
+ "834": [
2506
+ 6.16724967956543
2507
+ ],
2508
+ "835": [
2509
+ 5.965892124176025
2510
+ ],
2511
+ "836": [
2512
+ 5.974894046783447
2513
+ ],
2514
+ "837": [
2515
+ 6.383663336435954
2516
+ ],
2517
+ "838": [
2518
+ 6.217560172080994
2519
+ ],
2520
+ "839": [
2521
+ 6.189122486114502
2522
+ ],
2523
+ "840": [
2524
+ 6.376265335083008
2525
+ ],
2526
+ "841": [
2527
+ 6.5727012157440186
2528
+ ],
2529
+ "842": [
2530
+ 6.647420406341553
2531
+ ],
2532
+ "843": [
2533
+ 6.080006957054138
2534
+ ],
2535
+ "844": [
2536
+ 6.270581881205241
2537
+ ],
2538
+ "845": [
2539
+ 6.313574695587159
2540
+ ],
2541
+ "846": [
2542
+ 6.064110994338989
2543
+ ],
2544
+ "847": [
2545
+ 5.810511469841003
2546
+ ],
2547
+ "848": [
2548
+ 6.202533841133118
2549
+ ],
2550
+ "849": [
2551
+ 5.932036876678467
2552
+ ],
2553
+ "850": [
2554
+ 6.005986054738362
2555
+ ],
2556
+ "851": [
2557
+ 6.080901503562927
2558
+ ],
2559
+ "852": [
2560
+ 6.117233157157898
2561
+ ],
2562
+ "853": [
2563
+ 6.172588539123535
2564
+ ],
2565
+ "854": [
2566
+ 6.216827011108398
2567
+ ],
2568
+ "855": [
2569
+ 6.2789216836293535
2570
+ ],
2571
+ "856": [
2572
+ 6.059225463867188
2573
+ ],
2574
+ "857": [
2575
+ 6.413751761118571
2576
+ ],
2577
+ "858": [
2578
+ 5.949063777923584
2579
+ ],
2580
+ "859": [
2581
+ 6.295274666377476
2582
+ ],
2583
+ "860": [
2584
+ 6.388943791389465
2585
+ ],
2586
+ "861": [
2587
+ 6.217831543513706
2588
+ ],
2589
+ "862": [
2590
+ 6.239970684051514
2591
+ ],
2592
+ "863": [
2593
+ 5.958311939239502
2594
+ ],
2595
+ "864": [
2596
+ 6.242660045623779
2597
+ ],
2598
+ "865": [
2599
+ 6.541245555877685
2600
+ ],
2601
+ "866": [
2602
+ 5.9583148002624515
2603
+ ],
2604
+ "867": [
2605
+ 5.83608341217041
2606
+ ],
2607
+ "868": [
2608
+ 6.008572419484456
2609
+ ],
2610
+ "869": [
2611
+ 6.087687873840332
2612
+ ],
2613
+ "870": [
2614
+ 6.123212218284607
2615
+ ],
2616
+ "871": [
2617
+ 5.710743109385173
2618
+ ],
2619
+ "872": [
2620
+ 6.064690113067627
2621
  ]
2622
  },
2623
  "distributed/optimized-gpt2-250m-v0.1.1": {