Merge remote-tracking branch 'aav3-4/main' into jax

.gitignore

@@ -1,6 +1,49 @@
-docker
-postgres
-lost+found
-tmp
-/2023/07/06/introspector/root/
+# -*- mode: gitignore; -*-
 *~
+\#*\#
+/.emacs.desktop
+/.emacs.desktop.lock
+*.elc
+auto-save-list
+tramp
+.\#*
+
+# Org-mode
+.org-id-locations
+*_archive
+
+# flymake-mode
+*_flymake.*
+
+# eshell files
+/eshell/history
+/eshell/lastdir
+
+# elpa packages
+/elpa/
+
+# reftex files
+*.rel
+
+# AUCTeX auto folder
+/auto/
+
+# cask packages
+.cask/
+dist/
+
+# Flycheck
+flycheck_*.el
+
+# server auth directory
+/server/
+
+# projectiles files
+.projectile
+
+# directory configuration
+.dir-locals.el
+
+# network security
+/network-security.data
+

.gitmodules

@@ -3268,3 +3268,6 @@
 [submodule "2024/03/04/bumblebee"]
 	path = 2024/03/04/bumblebee
 	url = https://github.com/elixir-nx/bumblebee
+[submodule "2024/03/04/elixirconf2023"]
+	path = 2024/03/04/elixirconf2023
+	url = https://github.com/toranb/elixirconf2023.git

2024/02/setup.sh

@@ -9,3 +9,5 @@ asdf install elixir 1.12
 asdf global elixir 1.14.4-otp-26
 asdf install erlang 26.2.2
 asdf global erlang 26.2.2
+
+#export  LD_LIBRARY_PATH=/usr/local/lib/python3.10/dist-packages/libtpu

2024/03/03/notes.org

@@ -0,0 +1,7 @@
+the tpu oddessy is a strange one,
+the chip seems more for training than
+for serving inference.
+
+sudo apt install elixir cmake                                                sudo apt install libprotobuf-dev                                             sudo apt  install protobuf-compiler  # version                               sudo apt install libprotobuf-dev
+sudo apt install erlang-dev
+sudo apt install elixir cmake libprotobuf-dev protobuf-compiler erlang-dev

2024/03/04/notes.org

@@ -124,18 +124,66 @@ for the tpu odessy.
 funny that we found the elixir!
 
 
-
- 1001  git init --bare /mnt/data1/gittime
- 1002  git remote add local /mnt/data1/gittime
- 1003  git push local
- 1004  git branch
- 1005  git checkout -b 'jax'
- 1006  git push local
- 1007  git log -2
- 1008  git log -2 --all --patch
- 1009  git status
- 1010  git add 2024
- 1011  git commit -m 'addding' -a
- 1012  git push
- 1013  git push local
- 1014  history
+* setup of a git remote
+
+often we dont want to expose our keys to a external environment
+using ssh agents on remote hosts opens you up to having those keys stolen
+by anyone with root permissions on the host. better is to pull data via ssh back into your home back via ssh.
+in this example we will create a bare git repo on a remote server that
+we can access via a jump host, we will push to it and then pull back into our
+ home server.
+
+setup my .ssh/config
+
+#+begin_src .ssh/config
+host aaj2
+  HostName 123.45.78.90
+  User ubuntu
+
+Host aav3-4
+  HostName 10.164.0.21
+  User ubuntu	
+  ProxyJump  aaj2
+
+Host aav3-3
+  HostName 10.164.0.20
+  User ubuntu	
+  ProxyJump  aaj2
+
+Host aav3-2
+  HostName 10.164.0.19
+  User ubuntu	
+  ProxyJump  aaj2
+
+Host aav3-1
+  HostName 10.164.0.18
+  User ubuntu	
+  ProxyJump  aaj2
+#+end_src
+
+on the remote server construct a git time.
+
+#+begin_src shell
+ssh aav3-4
+#+end_src
+
+#+begin_src shell
+  git init --bare /mnt/data1/gittime
+  git remote add local /mnt/data1/gittime
+  git checkout -b 'example'
+  git push local
+#+end_src
+
+The contents of my git config in 
+/mnt/data1/nix/.git/modules/time/
+
+#+begin_src .git/config
+[remote "aav3-4"]
+	url = ssh://aav3-4:/mnt/data1/gittime
+	fetch = +refs/heads/*:refs/remotes/aav3-4/*
+#+end_src
+
+#+begin_src shell-history
+  git remote add aav3-4 ssh://aav3-4:/mnt/data1/gittime
+  git pull aav3-4 example
+#+end_src

test.py

@@ -0,0 +1,52 @@
+import torch_xla.core.xla_model as xm
+from transformers import AutoTokenizer, LlamaForCausalLM
+from human_eval.data import write_jsonl, read_problems
+from tqdm import tqdm
+
+# initialize the model
+
+#model_path = "Phind/Phind-CodeLlama-34B-v2"
+# Load model directly
+#model_path = "quastrinos/openbook-finetuned-deberta-v3-large-mcqa-TPU"
+model_path = 'openlm-research/open_llama_7b_v2'
+
+
+
+device = xm.xla_device()
+
+model = LlamaForCausalLM.from_pretrained(model_path, device_map="auto", )
+
+total_params = sum(p.numel() for p in model.parameters())
+print("Total number of parameters:", total_params)
+
+model.to(device)
+tokenizer = AutoTokenizer.from_pretrained(model_path).to(device)
+
+# HumanEval helper
+
+
+
+def generate_one_completion(prompt: str):
+    tokenizer.pad_token = tokenizer.eos_token
+    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=4096)
+
+
+    # Generate
+    generate_ids = model.generate(inputs.input_ids.to(device), max_new_tokens=384, do_sample=True, top_p=0.75, top_k=40, temperature=0.1)
+    completion = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+    completion = completion.replace(prompt, "").split("\n\n\n")[0]
+
+    return completion
+
+# perform HumanEval
+problems = read_problems()
+
+num_samples_per_task = 1
+samples = [
+    dict(task_id=task_id, completion=generate_one_completion(problems[task_id]["prompt"]))
+    for task_id in tqdm(problems)
+    for _ in range(num_samples_per_task)
+]
+write_jsonl("samples.jsonl", samples)
+
+# run `evaluate_functional_correctness samples.jsonl` in your HumanEval code sandbox