Update README.md
Browse files
README.md
CHANGED
@@ -136,45 +136,89 @@ print(outputs[0]["generated_text"][len(formatted_prompt):])
|
|
136 |
Training was done using QLoRA
|
137 |
|
138 |
|
139 |
-
Usage and Limitations
|
|
|
140 |
These models have certain limitations that users should be aware of.
|
141 |
|
142 |
-
Intended Usage
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
The
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
175 |
Risks identified and mitigations:
|
176 |
|
177 |
-
Perpetuation of biases: It's encouraged to perform continuous monitoring
|
178 |
-
|
179 |
-
|
180 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
Training was done using QLoRA
|
137 |
|
138 |
|
139 |
+
## Usage and Limitations
|
140 |
+
|
141 |
These models have certain limitations that users should be aware of.
|
142 |
|
143 |
+
### Intended Usage
|
144 |
+
|
145 |
+
Open Large Language Models (LLMs) have a wide range of applications across
|
146 |
+
various industries and domains. The following list of potential uses is not
|
147 |
+
comprehensive. The purpose of this list is to provide contextual information
|
148 |
+
about the possible use-cases that the model creators considered as part of model
|
149 |
+
training and development.
|
150 |
+
|
151 |
+
* Content Creation and Communication
|
152 |
+
* Text Generation: These models can be used to generate creative text formats
|
153 |
+
such as poems, scripts, code, marketing copy, and email drafts.
|
154 |
+
* Research and Education
|
155 |
+
* Natural Language Processing (NLP) Research: These models can serve as a
|
156 |
+
foundation for researchers to experiment with NLP techniques, develop
|
157 |
+
algorithms, and contribute to the advancement of the field.
|
158 |
+
* Language Learning Tools: Support interactive language learning experiences,
|
159 |
+
aiding in grammar correction or providing writing practice.
|
160 |
+
* Knowledge Exploration: Assist researchers in exploring large bodies of text
|
161 |
+
by generating summaries or answering questions about specific topics.
|
162 |
+
|
163 |
+
### Limitations
|
164 |
+
|
165 |
+
* Training Data
|
166 |
+
* The quality and diversity of the training data significantly influence the
|
167 |
+
model's capabilities. Biases or gaps in the training data can lead to
|
168 |
+
limitations in the model's responses.
|
169 |
+
* The scope of the training dataset determines the subject areas the model can
|
170 |
+
handle effectively.
|
171 |
+
* Context and Task Complexity
|
172 |
+
* LLMs are better at tasks that can be framed with clear prompts and
|
173 |
+
instructions. Open-ended or highly complex tasks might be challenging.
|
174 |
+
* A model's performance can be influenced by the amount of context provided
|
175 |
+
(longer context generally leads to better outputs, up to a certain point).
|
176 |
+
* Language Ambiguity and Nuance
|
177 |
+
* Natural language is inherently complex. LLMs might struggle to grasp subtle
|
178 |
+
nuances, sarcasm, or figurative language.
|
179 |
+
* Factual Accuracy
|
180 |
+
* LLMs generate responses based on information they learned from their
|
181 |
+
training datasets, but they are not knowledge bases. They may generate
|
182 |
+
incorrect or outdated factual statements.
|
183 |
+
* Common Sense
|
184 |
+
* LLMs rely on statistical patterns in language. They might lack the ability
|
185 |
+
to apply common sense reasoning in certain situations.
|
186 |
+
|
187 |
+
### Ethical Considerations and Risks
|
188 |
+
|
189 |
+
The development of large language models (LLMs) raises several ethical concerns.
|
190 |
+
In creating an open model, we have carefully considered the following:
|
191 |
+
|
192 |
+
* Bias and Fairness
|
193 |
+
* LLMs trained on large-scale, real-world text data can reflect socio-cultural
|
194 |
+
biases embedded in the training material. These models underwent careful
|
195 |
+
scrutiny, input data pre-processing described and posterior evaluations
|
196 |
+
reported in this card.
|
197 |
+
* Misinformation and Misuse
|
198 |
+
* LLMs can be misused to generate text that is false, misleading, or harmful.
|
199 |
+
* Guidelines are provided for responsible use with the model, see the
|
200 |
+
[Responsible Generative AI Toolkit](http://ai.google.dev/gemma/responsible).
|
201 |
+
* Transparency and Accountability:
|
202 |
+
* This model card summarizes details on the models' architecture,
|
203 |
+
capabilities, limitations, and evaluation processes.
|
204 |
+
* A responsibly developed open model offers the opportunity to share
|
205 |
+
innovation by making LLM technology accessible to developers and researchers
|
206 |
+
across the AI ecosystem.
|
207 |
+
|
208 |
Risks identified and mitigations:
|
209 |
|
210 |
+
* Perpetuation of biases: It's encouraged to perform continuous monitoring
|
211 |
+
(using evaluation metrics, human review) and the exploration of de-biasing
|
212 |
+
techniques during model training, fine-tuning, and other use cases.
|
213 |
+
* Generation of harmful content: Mechanisms and guidelines for content safety
|
214 |
+
are essential. Developers are encouraged to exercise caution and implement
|
215 |
+
appropriate content safety safeguards based on their specific product policies
|
216 |
+
and application use cases.
|
217 |
+
* Misuse for malicious purposes: Technical limitations and developer and
|
218 |
+
end-user education can help mitigate against malicious applications of LLMs.
|
219 |
+
Educational resources and reporting mechanisms for users to flag misuse are
|
220 |
+
provided. Prohibited uses of Gemma models are outlined in the
|
221 |
+
[Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy).
|
222 |
+
* Privacy violations: Models were trained on data filtered for removal of PII
|
223 |
+
(Personally Identifiable Information). Developers are encouraged to adhere to
|
224 |
+
privacy regulations with privacy-preserving techniques.
|