Let me first preface this with a few things. First, I'm not trying to promote anything. My models are free and open source and I'm not looking for attention. I genuinely love building and testing novel AI architectures, I have more model.py files lying around than I can count. If you feel the same and like to work on your own or other's AI models and test them out and try to train them, please reach out! From amateur hobbyists like myself to serious ML professionals, vibecoders and cracked pre-AI devs, all are welcome.
And secondly, I myself am technically a vibecoder. I try to keep the bar for my work as high as possible within that, quality wise, but I myself am still unable to manually write the code myself, and so fundamentally it's vibecoding. Though I do always check the output and iterate and run tests, code review, checking and formatting tools, the works. Thus far I've been able to build and train and test numerous small-mid sized custom novel architectures. Two of them are even open source on HuggingFace right now! The main one I'm working on right now is one of those two.
It's a completely novel architecture. It's called WrinkleBrane, and the idea is to use tensors as "membrane banks" using parallel read/write via orthogonal codes able to act as a memory layer for other AI, and using STE and parallel cumsum to act as its own standalone model. This is the one I want to work on right now.
I was able to train a 1.27M parameter, byte-level version of the standalone WrinkleBrane model over 40k steps on TinyStories and got it to below 2 perplexity from ~259 starting. Due to a bug causing a discrepancy between training and inference sampling, it only produces pseudo-english at <2 pplx, however if we were to fix the discrepancy we would likely see coherent TinyStories generation from this model at 1.27M parameters. It should also be noted that this model was trained entirely on the HuggingFace CPU upgrade space, so CPU only, 8 cores and 32GB RAM, no GPU or VRAM. It completed the 40k steps in about 20-24 total hours spread across two days, being 2 ~10hr 20k step runs total on the TinyStories corpus (plus some extra ascii tables and math and code I padded the dataset with)
The architecture has now gotten to the point where I need to get an outside perspective on it and get some collaborative work going. It's been public for a while, someone from Mem0 even contacted me about it in regards to WrinkleBrane's AI memory bank functionality. But then nothing. It hasn't gotten the attention it deserves yet, and it's further along than it's ever been!
Training and Sampling Snippets Below:
step 11950/20000 | loss=1.8341 ppl= 6.26 | lr=1.75e-02 gnorm=0.66 | 891 tok/s step 12000/20000 | loss=1.8401 ppl= 6.30 | lr=1.73e-02 gnorm=0.62 | 844 tok/s >>> VAL step 12000: loss=1.7827 ppl=5.95 ← best >>> Saved best_model.pt (loss=1.7827) --- sample @ step 12000 --- Once upon a time upos da a thexcon wex wa wagex. Shar(ved Jask Dat Sxcandd Lnacty Taramassnd Tolacenseasseeex "anlasonsex Tratas Olisateas "v --- step 12050/20000 | loss=1.8297 ppl= 6.23 | lr=1.72e-02 gnorm=0.67 | 774 tok/s step 12100/20000 | loss=1.8239 ppl= 6.20 | lr=1.70e-02 gnorm=0.74 | 857 tok/s step 13000/20000 | loss=1.7367 ppl= 5.68 | lr=1.37e-02 gnorm=0.74 | 826 tok/s >>> VAL step 13000: loss=1.6707 ppl=5.32 ← best >>> Saved best_model.pt (loss=1.6707) --- sample @ step 13000 --- Once upon a time, a yex a titirre, was was warean wamennd waittunndsh tandllllll waccckevea wttt warlllllllllllllllll Fll Iy T TOnenllllllernglan, W Clllllllll. Hlll --- step 13050/20000 | loss=1.7413 ppl= 5.70 | lr=1.35e-02 gnorm=0.84 | 724 tok/s step 14000/20000 | loss=1.6554 ppl= 5.24 | lr=1.04e-02 gnorm=0.83 | 1967 tok/s >>> VAL step 14000: loss=1.6069 ppl=4.99 ← best >>> Saved best_model.pt (loss=1.6069) --- sample @ step 14000 --- Once upon a time, a time, thereMas ais, wan was areann dllqBom wannnnnch wexsesnnnnntea annndsater. Tllllllllretlllllarr. Artllllllllll 'tllslllu O) --- step 14050/20000 | loss=1.6515 ppl= 5.21 | lr=1.02e-02 gnorm=0.87 | 1744 tok/s Entire Post-Training Inference Test Below:
Loading checkpoint: /data/WrinkleBrane-Research/09_standalone_model/checkpoints/best_model.pt Model: WrinkleBraneModelRoPE Step: 20000 Val loss: 1.4243330776691436 Parameters: 1,269,150 Config: d_model=128, n_layers=6, n_heads=4, L=16, K=128 ====================================================================== TEST 1: Story Prompts (TinyStories-style) ====================================================================== Prompt: "Once upon a time" Output: ", a a anyaFoice, a a wincerllll namamelllll wakkkllllll wlllllllllllllllllllllldlllllllllllrkkkllllllrrll.tllllllllaannnlllllllllllllllllrrlllleerllll" Prompt: "There was a little girl named" Output: "vooo n. anntimam ananain. lamanameman. sm lllssssememe w..w!"pe " wnslleeel....,,,e lle anneeeen,,,,?,,,,,,,,,, Ben Jl,*..":mee Haall ,, Eell I" Prompt: "The dog ran to the" Output: "amppen atit. her wanpppyanell w worsam. her bealeans sselean,.,,,,elererelll he terelsssl h..'releeeeenns.nl,,,,,,,,,eneee h sleer wlllllole,, "'lm,." Prompt: "One day, a boy found a" Output: "nd at Bevere woor d waneevv cov hox haneee s han sn hofeeallllllllslyyyyyy...hyyyyyyyyaneaeeaauy.t.beeeenreeeer,!.. Sylllllleieeeyen Seenlllllllll See" Prompt: "She was very happy because" Output: "t in smand way hin was sindd haitee, stt warooon wmoeain tt ttear. ufnaeais.! Ianllll Tinorlllleainen.eai<.....,yellllllleeeaaan, "t.,ellll..ellelener" Prompt: "The sun was shining and" Output: " thin her weelllle gaaaost etkk... Theeeeeaylll waiveeer w han sans..ieey heoax L hauss an w toommon wc Jinnn s hancsss has tseuss h his his-sssb" ====================================================================== TEST 2: Very Short Prompts ====================================================================== Prompt: "The" Output: "rer was whereat corlllesed, toom, an won wanmm hat hereas, ts hov, hereeays, jlllll T Shaiidtllleaielll D,,ssm. alllea" Prompt: "A" Output: "Acmy ana'D onY ans saix an wanay wat was want wan wat wadke hkeacc neet.. atc w wwons ann w. tofeeeeeed......'s T T "eee" Prompt: "He" Output: "r was woon andd was wan pux fellll han watcs han oxeamss, wssssseeat.." "aykllleeat. Hsss,.. T Barlllls Thaynne..glss.e" Prompt: "It" Output: " thereme fuicn boveain â Fit Benat, Him t horppeenaysssss bb w.h gonse sllknavss,dddvk,ddddddd,dsn.d.pnnd hnnvdddnndda" Prompt: "Hello" Output: "ust aneed anfow aneme, wan an sppps haisssunnnnnssssssssssssseeeakssstkss.wsaissnsss, n.o Wnnnlveaksnnarh. T/nllrseIyans" ====================================================================== TEST 3: Longer Prompts ====================================================================== Prompt: "Once upon a time, there was a little bear who lived in a big forest. One day, he decided to" Output: "k a lit theas wa weareapp faneat lllllllleeeeenn heel. wlllllllleeeearrlllllleadelllllllerelelllllcelllllelleell,lerenellelllllellelelee.e Bellelleren.' ulle 'lenerenel,,.'lllllealle,.,, W.e,l "" Prompt: "The princess looked out of the castle window and saw a dragon flying in the sky. She felt" Output: " fonelly, theay ateomed soailleeee Holyeol., Tooiiiin ieeannnsey,b w.in t ineor he wbb heoree h ppp dssll wcgs cppslepbbnfkdnvssdddssslsssendwg-kkvhvsskdaddsssssstddddwswsnn.ddsdlannddddsnnnnne.g" Prompt: "Mom said it was time to go to bed, but the little boy wanted to" Output: "om. this Hin sand. "Weporteret an.. M36, honnnt heeeee an...... Tddeeeen lls.. wreell... ww heeeeell heeeen, mb,,,,,,,,,,,,,'a .,,,,,,eeleeeeeeeel., heeeeeee, ', Ieeeeeeeeb,,,,, w ayeee '' 'W" ====================================================================== TEST 4: Math Prompts ====================================================================== Prompt: "2 + 3 =" Output: " 10 = 1161 52229 = = 11212984636707780515313978825" Prompt: "1 + 1 =" Output: " 100000 = = 11100 11 = = = 2 10 = 29999 0100000010" Prompt: "What is 5 + 7?" Output: "Yom is was wa sa ha sex souin c bin """" Tiimer.!." Prompt: "10 - 4 =" Output: " 14 1 = 110 = 8 009 = = 79169 8 = 1746778218994247" Prompt: "3 * 2 =" Output: " 4 370 = 3257 = = 66 34089723558275988748377366279" ====================================================================== TEST 5: Code Prompts ====================================================================== Prompt: "def hello" Output: "ved ant theear fof anns way han annnn wat an t waneay h. hat an tt hay tts wa.ay" Prompt: "for i in range" Output: "f lagiverlles wadeanttce wallllll haivv waitceaie wheaanllllleeeeeeeeeersseeer!." Prompt: "print(" Output: "asin and timm uponne Dinilll Heee abppleaesen, ayn ainn ainaxx..h Honnin aiss" Prompt: "if x > 0:" Output: " 8 268 = = = 1 462810229 = = 65387650689685689841778668887040eeeeeeeeeeeeeeeie" Prompt: "import " Output: "anadd and whin aneain t tarr ain alsseeeomen. Therllllllls. Liimchreannnll..... " ====================================================================== TEST 6: Temperature Comparison (same prompt, different temps) ====================================================================== [T=0.3] Prompt: "Once upon a time there was a" Output: " time, a a a a win winnttle wan wan llll wanllll wannnarsssseannelssssean. w. honnnntttsssssslllllll" [T=0.5] Prompt: "Once upon a time there was a" Output: " timeee, a an win wan wan wan wan wan wanassssesssesean. anean,s s, hexan wheans. Thexss.. "anlllll" [T=0.8] Prompt: "Once upon a time there was a" Output: "n anyme time anay, wanane wall wannnll wanasssscessesear. aneanns,s,,,,,,, lllllllllllls "B "Seallll" [T=1.0] Prompt: "Once upon a time there was a" Output: "n anyme time ane, lream wang wansnnnll wanassanc ssv tarsseneainsls,,,,,,, J' BoYNeayllss.B "Se"."'5" [T=1.5] Prompt: "Once upon a time there was a" Output: "n anyme hime smey, rismellegRl b2nnnll wvnadll c Cjv Taru Llkerltlslllea:j clllll.eTylls TB BSler.ػ" ====================================================================== TEST 7: Greedy Decoding (deterministic) ====================================================================== [greedy] Prompt: "Once upon a time" Output: ", a a a tima a a a winineallll waitlll wanllllllllllllllllllllllllllllllllllllllllllllllllllllllllll" [greedy] Prompt: "The little" Output: " girll garain ain an Timmmmmayyy. Tto That hat wat w wan w wan w wan w w wanppppppppppppennnnnn....e" [greedy] Prompt: "He was" Output: " a a snit was andeay ack anain t tanain tex. Thanex "ainsss.. "onnn w wan w won w wonllllllllleeeeee" ====================================================================== TEST 8: Top-k and Top-p Sampling ====================================================================== [top-k=10, T=0.8] Prompt: "The cat sat on the" Output: "ake garl ann theameaces tan cc lan bains,,,,,,,,, weme pmm wssserean bsmme wanlle wwwseeadd wnllell," [top-p=0.9, T=0.8] Prompt: "The cat sat on the" Output: "ake garl ann theameaces t torer wan wans, smsseaiman,sseme wssss,,..mkseme wappernnne wsadd wnlllll," [top-k=20, top-p=0.95, T=0.7] Prompt: "The cat sat on the" Output: "ake garl ann theameaces t torer wan wans, smsseaimel,sseme wssss,,..mkseme wappernnne wwadd wnllpll," ====================================================================== TEST 9: Next-Byte Prediction Analysis ====================================================================== Prompt: "The cat sat on the " Top predictions: 'b' : 0.1476 ####### 'p' : 0.1307 ###### 'w' : 0.1247 ###### 's' : 0.1158 ##### 'c' : 0.0709 ### 'g' : 0.0601 ### 'l' : 0.0466 ## 'f' : 0.0439 ## 'm' : 0.0386 # 'r' : 0.0368 # Prompt: "Once upon a " Top predictions: 't' : 0.9718 ################################################ 'n' : 0.0052 'i' : 0.0036 's' : 0.0033 'l' : 0.0027 'a' : 0.0018 ' ' : 0.0017 'b' : 0.0016 'd' : 0.0014 'f' : 0.0009 Prompt: "1 + 1 = " Top predictions: '1' : 0.3199 ############### '0' : 0.1789 ######## 'r' : 0.0721 ### '2' : 0.0626 ### '3' : 0.0495 ## '4' : 0.0397 # '5' : 0.0366 # '6' : 0.0355 # '7' : 0.0311 # ' ' : 0.0298 # Prompt: "She was very " Top predictions: 'w' : 0.1499 ####### 'a' : 0.1138 ##### 's' : 0.0905 #### 'm' : 0.0803 #### 't' : 0.0786 ### 'h' : 0.0756 ### 'f' : 0.0711 ### 'b' : 0.0505 ## 'c' : 0.0444 ## 'p' : 0.0363 # Prompt: "Hello " Top predictions: 'w' : 0.1843 ######### 'a' : 0.1133 ##### 'g' : 0.1042 ##### 's' : 0.0786 ### 'l' : 0.0733 ### 'd' : 0.0494 ## 'o' : 0.0473 ## 'b' : 0.0460 ## 'c' : 0.0450 ## 't' : 0.0445 ## ====================================================================== TEST 10: Perplexity on Sample Texts ====================================================================== [GOOD] PPL= 1.56 Simple story: "Once upon a time there was a little cat." [GOOD] PPL= 5.55 Common English: "The dog is big and brown." [GOOD] PPL= 5.13 Repetitive: "the the the the the the" [HIGH] PPL= 898.63 Nonsense: "xqz pfk wvn mrl bgt" [OK ] PPL= 44.48 Numbers: "1 2 3 4 5 6 7 8 9 10" [GOOD] PPL= 2.34 Story-like: "The little girl smiled and ran to the park." [HIGH] PPL= 116.94 Code-like: "def foo(x): return x + 1" ====================================================================== TEST 11: Determinism / Consistency Check ====================================================================== Prompt: "The bear walked into" Run 1: "gher toosy fin ain anddrrencssleie... Gsanllll..." Run 2: "gher toosy fin ain anddrrencssleie... Gsanllll..." Run 3: "gher toosy fin ain anddrrencssleie... Gsanllll..." All identical: True PASS ====================================================================== TEST 12: Generation Speed ====================================================================== 50 tokens: 4.48s (11.2 tok/s) 100 tokens: 7.51s (13.3 tok/s) 200 tokens: 13.10s (15.3 tok/s) ====================================================================== TEST 13: Edge Cases ====================================================================== --- Single character prompt --- [single char] Prompt: "a" Output: "re was theered for oface wan, wanayss heass, w sanppppeay un hs anssssss tonn.h " --- Punctuation prompt --- [punctuation] Prompt: "..." Output: " Hee The s asa wanean is mopombbeweeeeeeeour., II W OM'mlllllle1ill Hallllllleay" --- Numeric prompt --- [numbers] Prompt: "12345" Output: " = 3886 72289967753318938576966441422909870emav117625mftttleeees = 3220830995052" --- Newline in prompt --- [newlines] Prompt: "Hello\nWorld\n" Output: "Hello World keeepped arrrserlerederersenthin..! T Sher sean acc innnnn Jppewsssl han anasss" ====================================================================== TEST 14: Continuation Coherence (does it follow the prompt?) ====================================================================== [Emotional] Prompt: "The boy was sad because his toy was broken. He started to" Output: " an trame win wa fanex hin hanee honsss heeellllllllllllllllllllleanrllllllllllllllllldpsllllllrllllleorllllllllllllllll" [Action] Prompt: "The bird flew up high into the" Output: "nrer speeair. finaine f hinear has. hin haneeee t sannssseellll ssseleemmmanslnddnnsssssssll hetmssskennnkevd annnnnnnnl" [Dialogue] Prompt: ""Can I have some cake?" asked the little girl. Mom said, "" Output: " am amattty ack yoove taccy, tov wanovv fanov wainead. M f ballllexeaid.,,, "earead.. Alllllll!ere "on,,,,dereleaerll "e" [Descriptive] Prompt: "The garden was full of beautiful flowers. There were red roses and" Output: "ay a lsppppy, herteier ar teamns t. On Theeeeeesll J. inn,,"nneeay wwannnn, Tye hsnle hnssssleennnn. w annnkensandnddd" ====================================================================== TEST 15: Repetition Penalty (key test for char repetition) ====================================================================== [rep_pen=1.0] Prompt: "Once upon a time there was a" Output: " time, a a a a wititt walll waname w anll wanax wanannnesss.. wanev f llexan. hexex, wllllsssslllllllllerlllllearlllller" [rep_pen=1.1] Prompt: "Once upon a time there was a" Output: "ntime, a a anitime, was leamel wannnnag annnessssessssesssk.. ts sainnlleain...,,,,,,,,'llllllllllla '' Leoaillllllllep" [rep_pen=1.2] Prompt: "Once upon a time there was a" Output: "ntime, a a anitime, walll wanllem dat wan wanssssait. wssskeannanev fslleain...,,,,,,,,'lllllearoneal,,,.,,,eeeereeeleep" [rep_pen=1.5] Prompt: "Once upon a time there was a" Output: "nt, ame, a a lifeaFh was wanellly dag wlearkessh. bannnev kexsss ev falleain..,, ex' wllllsslearonualleerlerere H 1ealep" [rep_pen=2.0] Prompt: "Once upon a time there was a" Output: "ny, amex." ====================================================================== TEST 16: Min-p Sampling ====================================================================== [min_p=0.0] Prompt: "The little girl walked to the park and" Output: "nvim Onnn rever tFripoil t. S6, SSmmmmee 'mm Sheme ame udk." Taye immmme an,sssseme hesssssssssonuasssssunkssssssssandp" [min_p=0.02] Prompt: "The little girl walked to the park and" Output: "nt gimmmmeeremee t.. Emmmerrreremend. Alerennssssein.. On Th Tm wev inlleanadmad ws hlllmssssannnnas,ses,b,b,ennan,,wep" [min_p=0.05] Prompt: "The little girl walked to the park and" Output: "nt gimmmmeeremee t.. Emmmerrreremend. Alerennssssein.. On Th Tm weaiinlle an,,,,,,,,,,,,,,,,ye Tannas, h w an ww an an w" [min_p=0.1] Prompt: "The little girl walked to the park and" Output: "nt gimmmmeeremee t.. Onn Timmm Onmmme Ammay, Sx, anamme ud, Janay, Timmme ay, am L S w s an w wonn ayppps w ane anpsan.." ====================================================================== TEST 17: Combined Strategy Comparison ====================================================================== [Baseline T=0.8] Prompt: "Once upon a time, there was a little bear who lived in a forest." Output: " therere war wanepppin s winelleamereail. 'deayereerereereererearererelleraiiyeenerereerleeon. ronlayenerlererenqhellepllee hldlan. weallellllll+ Bee" [Low T + top-k] Prompt: "Once upon a time, there was a little bear who lived in a forest." Output: " therere war wanean wa wa werereeee teeeeererererererenleererenllererellerarenlllererlllllleeenrenlarenlllenllenllllllerenlenll, wllenlellenlllllle he" [Top-p=0.9 + rep_pen=1.2] Prompt: "Once upon a time, there was a little bear who lived in a forest." Output: " therere war wanepppin s wineaneee weay,, Aneayeeelerenereererearereerlleeaiiyellererlllllleeearenlllllerlereeenllllleerelle hlelen. w,,,,,?"llllee an" [Min-p=0.05 + rep_pen=1.3] Prompt: "Once upon a time, there was a little bear who lived in a forest." Output: "" yence wan wa wa lis haleamel weean aineereedereain.. Oudeearearererellereeeneenererlllllleeellennaid L.,,,ailllllllerelele Jl,,an. S,,,llllllllle an" [Top-k=30 + min-p=0.02 + rep_pen=1.2] Prompt: "Once upon a time, there was a little bear who lived in a forest." Output: "" uponn a areve war wous warelleeeereail. Adeayeererereereer.!eareeeerlleeneeneerererllllllelelennlelllerlerelenllllleerelle hl, wlleennllellllllen he" [Aggressive anti-repeat] Prompt: "Once upon a time, there was a little bear who lived in a forest." Output: "" yence wan wa wa lis halpppeleeeel! aineereedererer.. Auiierenlleeerelleeneeneerererllllllllellenen.'nerlererelllllllereleeall, Sl. hellellllllee Ee" ====================================================================== TEST 18: Parallel vs Sequential Mode Agreement ====================================================================== Text: "Once upon a time there was a cat" Max logit difference: 1.15e+00 Mean logit difference: 1.41e-01 Status: CHECK ====================================================================== TEST 19: Multi-Pass Iterative Refinement ====================================================================== Concept: Feed the model's own output back as input for another generation pass. Tests if iterative self-refinement improves coherence (leveraging membrane state accumulation). Pass 1: " uponday, wanns sanaman w wanain, annain ain Bomeeeeeereeeors.. THaearereelerel" Pass 2: "owar hammmmmmee anddee a a himm he amm Soyexoo toousasommay, Shayo" Pass 3: " eeele s IWWYCMaxaman,,""",,,'manevvan,,,,!inex v,,,veuns b, cexxan w leon " Pass 4: "ouuuueel ccccaneave wlllllkeaaackeeeew aw A Jeeeeeeeeeeeeeeeee Allllllle" Final accumulated text (360 chars): "Once upon a time there was a little cat. uponday, wanns sanaman w wanain, annain ain Bomeeeeeereeeors.. THaearereelerelowar hammmmmmee anddee a a himm he amm Soyexoo toousasommay, Shayo eeele s IWWYCMaxaman,,""",,,'manevvan,,,,!inex v,,,veuns b, cexxan w leon ouuuueel ccccaneave wlllllkeaaackeeeew aw A Jeeeeeeeeeeeeeeeee Allllllle" ====================================================================== TEST 20: Token-level Entropy Analysis ====================================================================== Per-byte entropy and confidence for: "Once upon a time there was a little girl named Lily." Byte Char Entropy MaxProb Conf ---------------------------------------- 0 BOS 1.009 0.8308 HIGH 1 O 0.260 0.9667 HIGH 2 n 0.390 0.9244 HIGH 3 c 0.134 0.9819 HIGH 4 e 0.240 0.9595 HIGH 5 0.434 0.9281 HIGH 6 u 0.065 0.9927 HIGH 7 p 0.163 0.9766 HIGH 8 o 0.085 0.9902 HIGH 9 n 0.142 0.9796 HIGH 10 0.110 0.9857 HIGH 11 a 0.054 0.9940 HIGH 12 0.208 0.9718 HIGH 13 t 0.165 0.9763 HIGH 14 i 0.080 0.9903 HIGH 15 m 0.055 0.9935 HIGH 16 e 0.551 0.8612 HIGH 17 0.648 0.8773 HIGH 18 t 0.586 0.8600 HIGH 19 h 0.173 0.9713 HIGH 20 e 0.170 0.9766 HIGH 21 r 0.068 0.9914 HIGH 22 e 0.263 0.9588 HIGH 23 1.155 0.7807 HIGH 24 w 0.770 0.8107 HIGH 25 a 0.338 0.9410 HIGH 26 s 0.062 0.9928 HIGH 27 0.462 0.9288 HIGH 28 a 0.349 0.9409 HIGH 29 1.988 0.4318 MED 30 l 0.435 0.9047 HIGH 31 i 1.021 0.7910 HIGH 32 t 0.046 0.9945 HIGH 33 t 0.091 0.9875 HIGH 34 l 0.099 0.9863 HIGH 35 e 0.120 0.9839 HIGH 36 0.936 0.7542 HIGH 37 g 0.500 0.8965 HIGH 38 i 0.193 0.9735 HIGH 39 r 0.089 0.9890 HIGH 40 l 0.474 0.9088 HIGH 41 1.464 0.6898 HIGH 42 n 0.333 0.9410 HIGH 43 a 0.276 0.9611 HIGH 44 m 0.219 0.9641 HIGH 45 e 0.351 0.9381 HIGH 46 d 0.830 0.7953 HIGH 47 2.404 0.2786 MED 48 L 0.457 0.9105 HIGH 49 i 0.072 0.9911 HIGH 50 l 0.119 0.9816 HIGH 51 y 0.992 0.6368 HIGH 52 . 0.364 0.9191 HIGH Average entropy: 0.435 Average max prob: 0.9046 (Lower entropy = more confident, higher max prob = more peaked) ====================================================================== SMOKETEST COMPLETE ====================================================================== All 20 tests executed. The WrinkleBrane model is loaded and generating text. [link] [comments]