| BOUNDING_BOX_TOOL_NAME = "capture_bounding_boxes" | |
| BOUNDING_BOX_TOOL = { | |
| "type": "function", | |
| "name": BOUNDING_BOX_TOOL_NAME, | |
| "strict": True, | |
| "description": ( | |
| "Capture normalized bounding boxes for salient regions in the provided image. " | |
| "Each box should map to a meaningful real-world entity or text block and use clockwise point order." | |
| ), | |
| "parameters": { |
| 以下是该处方笺的自然阅读内容，已根据手写体进行识别和整理： | |
| --- | |
| **处方笺** | |
| **日期：** 14.6.6号 | |
| **姓名：** [已涂黑] | |
| **性别：** 女 | |
| **年龄：** 24 |
| Groups | Version | Filter | n-shot | Metric | Value | Stderr | ||
|---|---|---|---|---|---|---|---|---|
| leaderboard | N/A | none | 0 | acc | ↑ | 0.4464 | ± | 0.0045 |
| none | 0 | acc_norm | ↑ | 0.5778 | ± | 0.0051 | ||
| none | 0 | exact_match | ↑ | 0.1992 | ± | 0.0055 | ||
| none | 0 | inst_level_loose_acc | ↑ | 0.8297 | ± | N/A | ||
| none | 0 | inst_level_strict_acc | ↑ | 0.8058 | ± | N/A | ||
| none | 0 | prompt_level_loose_acc | ↑ | 0.7671 | ± | 0.0182 | ||
| none | 0 | prompt_level_strict_acc | ↑ | 0.7338 | ± | 0.0190 | ||
| - leaderboard_bbh | N/A | none | 3 | acc_norm | ↑ | 0.6454 | ± | 0.0057 |
| Mapping citations | |
| First, I’m identifying each citation within the text, matching them to the Documents section. This involves noting the citation format and ensuring relevancy to the sentence it supports. | |
| Cross-checking citations | |
| I’m piecing together how the RESPONSE references documents like “According to Document 3” and “Document 9 highlights the importance of psychological flexibility” to ensure the citations accurately reflect the content in the DOCUMENTS. | |
| Verifying citations | |
| Interestingly enough, I’m checking the accuracy of each citation in the RESPONSE by cross-referencing them with the REFERENCES section. This guarantees precise alignment with the document. | |
| Verifying citations | |
| The assistant is instructed to scrutinize citations in the RESPONSE for relevance, cross-reference them with the REFERENCES section, and confirm their accuracy. | |
| ACTIONS NEEDED: | |
| I will read the RESPONSE paragraph by paragraph, assessing citations for appropriateness. Summarizing content may be necessary due to length. |
| base_model: /workspace/models/Mistral-Nemo-Base-2407 | |
| model_type: AutoModelForCausalLM | |
| tokenizer_type: AutoTokenizer | |
| load_in_8bit: false | |
| # load_in_4bit: true | |
| strict: false | |
| datasets: | |
| - path: /workspace/datasets/dolphin-2.9.3/dolphin201-sharegpt2.jsonl |
| File "/root/miniconda3/envs/py3.10/lib/python3.10/site-packages/trl/trainer/utils.py", line 338, in __call__ | |
| to_pad = [torch.LongTensor(ex[k]) for ex in features] | |
| File "/root/miniconda3/envs/py3.10/lib/python3.10/site-packages/trl/trainer/utils.py", line 338, in <listcomp> | |
| to_pad = [torch.LongTensor(ex[k]) for ex in features] | |
| File "/root/miniconda3/envs/py3.10/lib/python3.10/site-packages/trl/trainer/utils.py", line 338, in <listcomp> | |
| to_pad = [torch.LongTensor(ex[k]) for ex in features] | |
| TypeError: 'NoneType' object cannot be interpreted as an integer | |
| to_pad = [torch.LongTensor(ex[k]) for ex in features] | |
| TypeError: 'NoneType' object cannot be interpreted as an integer | |
| return inner_training_loop( |
Let's checkout the PR:
git fetch origin pull/625/head:dbrx
git switch dbrx
pip install -vvv --no-build-isolation -e .Download the model:
| from transformers import LlamaConfig as LC, LlamaForCausalLM as LLM, LlamaTokenizer as LT | |
| from accelerate import init_empty_weights, load_checkpoint_and_dispatch | |
| import torch | |
| lt = LT.from_pretrained("NousResearch/Llama-2-7b-hf") | |
| c = LC.from_pretrained("NousResearch/Llama-2-70b-hf") | |
| c.max_position_embeddings = 32764 | |
| c.rope_theta = 1000000 | |
| with init_empty_weights(): m = LLM(c) | |
| m = m.half().eval() | |
| m.requires_grad_(False) |
| #!/bin/sh | |
| log() { | |
| format="$1"; shift | |
| # shellcheck disable=SC2059 | |
| >&2 printf "$format\n" "$@" | |
| } | |
| usage() { | |
| >&2 cat <<EOF |