Upload 13 files

models/retriever/level-4-small-no-negative-outcomes/document_index/config.yaml

@@ -0,0 +1,5 @@
+_target_: relik.retriever.indexers.inmemory.InMemoryDocumentIndex
+metadata_fields:
+- definition
+separator: ' <def> '
+name_or_path: null

models/retriever/level-4-small-no-negative-outcomes/document_index/embeddings.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f48d0753bec8ae343e33355d98538895384ee9115d421e625f22420164a7c5bf
+size 1352875

models/retriever/level-4-small-no-negative-outcomes/document_index_db/config.yaml

@@ -0,0 +1,5 @@
+_target_: relik.retriever.indexers.inmemory.InMemoryDocumentIndex
+metadata_fields:
+- definition
+separator: ' <def> '
+name_or_path: null

models/retriever/level-4-small-no-negative-outcomes/document_index_db/documents.jsonl

@@ -0,0 +1,79 @@
+{"text": "marriage and teen child bearing as a barrier to education participation", "id": 0, "metadata": {"definition": "outcome. measures of marriage and teenage child bearing in the local context, which may act as a barrier to participation in education.", "level": 4, "type": "outcome"}}
+{"text": "business knowledge", "id": 1, "metadata": {"definition": "outcome. any measure of business practices such as sharing business experiences, asking for business advice and interest in training. this may also include the understanding and ability to apply business concepts in real-life situation such as creating a business plan, understanding how to calculate profit etc.", "level": 4, "type": "outcome"}}
+{"text": "wage levels", "id": 2, "metadata": {"definition": "outcome. measures for wage levels or wage level satisfaction", "level": 4, "type": "outcome"}}
+{"text": "investment in productive assets", "id": 3, "metadata": {"definition": "outcome. total amount of investments made on equipment for productive purposes.", "level": 4, "type": "outcome"}}
+{"text": "index of illness related to wash", "id": 4, "metadata": {"definition": "outcome. composite/index measures of other health indicators not otherwise captured related to wash topics.", "level": 4, "type": "outcome"}}
+{"text": "index of healthcare professional outcomes", "id": 5, "metadata": {"definition": "outcome. composite/index measures of health care professional staff availability, quality, and supervision.", "level": 4, "type": "outcome"}}
+{"text": "profits and revenues", "id": 6, "metadata": {"definition": "outcome. amount of profit or revenues generated by a firm or business over a given period.", "level": 4, "type": "outcome"}}
+{"text": "gendered access to credit", "id": 7, "metadata": {"definition": "outcome. women's access to credit through formal institutions or informal lenders.", "level": 4, "type": "outcome"}}
+{"text": "job seeking", "id": 8, "metadata": {"definition": "outcome. measures of job seeking behaviour", "level": 4, "type": "outcome"}}
+{"text": "attitudes about intimate partner violence", "id": 9, "metadata": {"definition": "outcome. measures of attitudes with respect to intimate partner violence, including identity formation, perception of gender roles, acceptability of sexist attitudes, acceptability of ipv, intimacy and self-efficacy. this indicator is applicable at an individual, family, or community level.", "level": 4, "type": "outcome"}}
+{"text": "index of health service quality & standards", "id": 10, "metadata": {"definition": "outcome. composite/index measures of health service quality and standards.", "level": 4, "type": "outcome"}}
+{"text": "unemployment", "id": 11, "metadata": {"definition": "outcome. measures of unemployment status.", "level": 4, "type": "outcome"}}
+{"text": "gender representation", "id": 12, "metadata": {"definition": "outcome. measures of representation by gender to improve access to services and increase social mobility, including reservations.", "level": 4, "type": "outcome"}}
+{"text": "index of sanitation attitudes and knowledge", "id": 13, "metadata": {"definition": "outcome. composite/index measures of attitudes towards sanitation practices, and all measures of knowledge about sanitation practices and related topics (such as disease transmission).", "level": 4, "type": "outcome"}}
+{"text": "cattle herd size", "id": 14, "metadata": {"definition": "outcome. overall cattle holdings for a farm/ranch.", "level": 4, "type": "outcome"}}
+{"text": "index of household income", "id": 15, "metadata": {"definition": "outcome. composite/index measures of household income.", "level": 4, "type": "outcome"}}
+{"text": "empowerment", "id": 16, "metadata": {"definition": "outcome. process through which someone acquires the ability to access certain resources, make independent choices based on his/ her own values, and achieve things that correspond to his/her own values and objectives. this can be measured either through the observation of a set of behaviours that demonstrate an increased access to resources, agency and achievements, or through self-reported perception of someone's own empowerment.", "level": 4, "type": "outcome"}}
+{"text": "social participation and interaction", "id": 17, "metadata": {"definition": "outcome. socialisation and interaction measures.", "level": 4, "type": "outcome"}}
+{"text": "overall household income", "id": 18, "metadata": {"definition": "outcome. total household income from all income-generating activities.", "level": 4, "type": "outcome"}}
+{"text": "index of hygiene attitudes and knowledge", "id": 19, "metadata": {"definition": "outcome. composite/index measures of attitudes towards hygiene practices, and all measures of knowledge about hygiene practices and related topics.", "level": 4, "type": "outcome"}}
+{"text": "sexual and intimate partner violence", "id": 20, "metadata": {"definition": "outcome. measures of sexual and intimate partner violence incidence", "level": 4, "type": "outcome"}}
+{"text": "index of corruption", "id": 21, "metadata": {"definition": "outcome. composite/index measures of corruption.", "level": 4, "type": "outcome"}}
+{"text": "handwashing behaviour - before handling food", "id": 22, "metadata": {"definition": "outcome. all measures of how consistently participants practice the handwashing behaviours that they are taught. only in relation to food preparation activities.", "level": 4, "type": "outcome"}}
+{"text": "index of business knowledge and skills", "id": 23, "metadata": {"definition": "outcome. composite/index measures of business knowledge and skills.", "level": 4, "type": "outcome"}}
+{"text": "tree replacement", "id": 24, "metadata": {"definition": "outcome. measures of amount and quality of tree replacement programs, and recovered timberland.", "level": 4, "type": "outcome"}}
+{"text": "awareness of rights", "id": 25, "metadata": {"definition": "outcome. knowledge and understanding of one's political, civil, and human rights.", "level": 4, "type": "outcome"}}
+{"text": "perceptions of personal and community safety", "id": 26, "metadata": {"definition": "outcome. this includes measures of perceived safety in the home, safety in the community and freedom of movement.", "level": 4, "type": "outcome"}}
+{"text": "tax compliance", "id": 27, "metadata": {"definition": "outcome. measures of tax compliance. these include measures of tax evasion and the amount of taxes paid.", "level": 4, "type": "outcome"}}
+{"text": "employment status", "id": 28, "metadata": {"definition": "outcome. type of employment status under which a person is recognize by the law, determining his/her rights and the employers' responsibility. the main employment status are self-employed or contractor, director, worker, employee, unemployed, retired.", "level": 4, "type": "outcome"}}
+{"text": "diarrhoea-related mortality", "id": 29, "metadata": {"definition": "outcome. any measure of the number of deaths from diarrhoeal diseases.", "level": 4, "type": "outcome"}}
+{"text": "access to public education", "id": 30, "metadata": {"definition": "outcome. any measure of the availability and capacity of public education services, including opening hours, waiting times and equitable access to services and resources made available by service providers.", "level": 4, "type": "outcome"}}
+{"text": "index of social cohesion", "id": 31, "metadata": {"definition": "outcome. composite/index of measures of social cohesion within a community.", "level": 4, "type": "outcome"}}
+{"text": "migration", "id": 32, "metadata": {"definition": "outcome. any measure of the amount of migration in our out of the area (usually a headcount).", "level": 4, "type": "outcome"}}
+{"text": "sales", "id": 33, "metadata": {"definition": "outcome. volume of goods or services sold by a firm.", "level": 4, "type": "outcome"}}
+{"text": "index of latrine or toilet use", "id": 34, "metadata": {"definition": "outcome. composite/index measures of the latrine or toilet use.", "level": 4, "type": "outcome"}}
+{"text": "government expenditure", "id": 35, "metadata": {"definition": "outcome. total amount spent by the government", "level": 4, "type": "outcome"}}
+{"text": "index of faecal waste disposal", "id": 36, "metadata": {"definition": "outcome. composite/index measures of waste disposal methods.", "level": 4, "type": "outcome"}}
+{"text": "regular latrine or toilet use", "id": 37, "metadata": {"definition": "outcome. all measures of how regularly participants use latrine or toilets.", "level": 4, "type": "outcome"}}
+{"text": "sanitation-related mortality", "id": 38, "metadata": {"definition": "outcome. any measure of the number deaths from sanitation-related diseases.", "level": 4, "type": "outcome"}}
+{"text": "open defecation", "id": 39, "metadata": {"definition": "outcome. all measures of how consistently participants openly defecate, or what proportion of participants who no longer openly defecate at all.", "level": 4, "type": "outcome"}}
+{"text": "safe disposal of waste", "id": 40, "metadata": {"definition": "outcome. all measures of the number, or proportion, of participants who consistently participants practice the safe disposal of faeces.", "level": 4, "type": "outcome"}}
+{"text": "new latrine or toilet construction", "id": 41, "metadata": {"definition": "outcome. number of new latrine or toilets constructed over a given period.", "level": 4, "type": "outcome"}}
+{"text": "adoption of good business practices from higher education", "id": 42, "metadata": {"definition": "outcome. measures of the adoption of business practices associated with good business outcomes. these practices may include regular bookkeeping, clear recordkeeping, good planning, and other practices that are taught during business trainings. we included studies that measured the adoption of these and other good business practices.", "level": 4, "type": "outcome"}}
+{"text": "income from agriculture", "id": 43, "metadata": {"definition": "outcome. income stemming from all forms of agricultural production.", "level": 4, "type": "outcome"}}
+{"text": "social norms around sanitation", "id": 44, "metadata": {"definition": "outcome. all measures of the perceived social norms around sanitation practices.", "level": 4, "type": "outcome"}}
+{"text": "community monitoring", "id": 45, "metadata": {"definition": "outcome. participation in community monitoring of public resources and projects.", "level": 4, "type": "outcome"}}
+{"text": "index of social attitudes, behaviours, or norms.", "id": 46, "metadata": {"definition": "outcome. composite/index of measures of social attitudes, behaviours, or norms.", "level": 4, "type": "outcome"}}
+{"text": "women's employment after higher education", "id": 47, "metadata": {"definition": "outcome. womens employment status and type of occupation, including whether women were employed, the status of their employment (full-time/part-time, permanent/temporary, formal/informal), and the type of occupation in which they were employed.", "level": 4, "type": "outcome"}}
+{"text": "access to healthcare", "id": 48, "metadata": {"definition": "outcome. measures of access, physical and financial, to healthcare services.", "level": 4, "type": "outcome"}}
+{"text": "index of sexual & reproductive health", "id": 49, "metadata": {"definition": "outcome. composite/index measures of sexual and reproductive health outcomes and services.", "level": 4, "type": "outcome"}}
+{"text": "customer satisfaction", "id": 50, "metadata": {"definition": "outcome. customer satisfaction measures for a product or service.", "level": 4, "type": "outcome"}}
+{"text": "perception of politicians' responsiveness", "id": 51, "metadata": {"definition": "outcome. public perception of politicians' responsiveness to constituents concerns and accessibility. this can include constituencies' feelings of being heard by leaders.", "level": 4, "type": "outcome"}}
+{"text": "attitudes toward open defecation", "id": 52, "metadata": {"definition": "outcome. attitudes toward open defecation, including levels of acceptance.", "level": 4, "type": "outcome"}}
+{"text": "customer retention", "id": 53, "metadata": {"definition": "outcome. measures of customer retention, including returning or continuous customers.", "level": 4, "type": "outcome"}}
+{"text": "index of firm behaviour", "id": 54, "metadata": {"definition": "outcome. composite/index measures related to firm behaviour and performance regarding generalised topics of public concern.", "level": 4, "type": "outcome"}}
+{"text": "household expenditure on food", "id": 55, "metadata": {"definition": "outcome. amount or portion of expenditure on by the household.", "level": 4, "type": "outcome"}}
+{"text": "perception of public service delivery", "id": 56, "metadata": {"definition": "outcome. public perception of service quality and accessibility.", "level": 4, "type": "outcome"}}
+{"text": "handwashing behaviour", "id": 57, "metadata": {"definition": "outcome. all measures of how consistently participants practice the handwashing behaviours that they are taught.", "level": 4, "type": "outcome"}}
+{"text": "social norms regarding violence", "id": 58, "metadata": {"definition": "outcome. this includes norms and behaviours surrounding violence, including support for political violence or armed groups, and attitudes towards the use of violence", "level": 4, "type": "outcome"}}
+{"text": "women's earning/income after higher education", "id": 59, "metadata": {"definition": "outcome. measures of income women earned through paid work, defined as earnings, wages, or salaries. measures of earnings or wages through administrative data, as well as self-reported data where women had been asked about their earnings, wages, or salaries.", "level": 4, "type": "outcome"}}
+{"text": "knowledge of handwashing practices", "id": 60, "metadata": {"definition": "outcome. knowledge of handwashing practices.", "level": 4, "type": "outcome"}}
+{"text": "social trust", "id": 61, "metadata": {"definition": "outcome. any measure of trust between communities, their members, and across different social groups. it may include specific measures of respondents' perception of others', trust towards each other and tolerance.", "level": 4, "type": "outcome"}}
+{"text": "acute respiratory infections (aris)", "id": 62, "metadata": {"definition": "outcome. any measure of acute illness in the upper respiratory tract, trachea, bronchi, bronchioles, alveoli, pleura, pleural cavity, and the muscles associated with breathing", "level": 4, "type": "outcome"}}
+{"text": "wash all-cause mortality", "id": 63, "metadata": {"definition": "outcome. any measure of the number of deaths from any cause.", "level": 4, "type": "outcome"}}
+{"text": "food insecurity prevalence", "id": 64, "metadata": {"definition": "outcome. measures of food insecurity in surrounding communities and within the household.", "level": 4, "type": "outcome"}}
+{"text": "diarrhoeal disease", "id": 65, "metadata": {"definition": "outcome. all measures of diarrhoeal disease. all reasonable definitions of diarrhoea accepted (frequent, loose-watery stool). this is typically presented as number of incidences over a given period.", "level": 4, "type": "outcome"}}
+{"text": "employee motivation", "id": 66, "metadata": {"definition": "outcome. any measure of employee motivation or the firm's ability to motivate employees.", "level": 4, "type": "outcome"}}
+{"text": "employment share", "id": 67, "metadata": {"definition": "outcome. portion of the labour force in a particular sector.", "level": 4, "type": "outcome"}}
+{"text": "index of empowerment", "id": 68, "metadata": {"definition": "outcome. composite/index measures of social or economic empowerment", "level": 4, "type": "outcome"}}
+{"text": "participation and inclusion", "id": 69, "metadata": {"definition": "outcome. measures of participation and inclusion in social, business, and political activities.", "level": 4, "type": "outcome"}}
+{"text": "resilience", "id": 70, "metadata": {"definition": "outcome. resilience is the ability to manage risks and recover from shocks, economically and psychologically, and can be measured either through observing a set of behaviours which demonstrate resilience or it can be a self-reported perception of resilience. the set of behaviours which demonstrates resilience are for example the diversification of business activity to manage risk, the strategies adopted to cope with unpredictable events such as a drought or an increase in input prices, or even mental health state after a loss.", "level": 4, "type": "outcome"}}
+{"text": "self-employment earnings due to higher education", "id": 71, "metadata": {"definition": "outcome. self-employment earnings due to higher education.", "level": 4, "type": "outcome"}}
+{"text": "business formalisation", "id": 72, "metadata": {"definition": "outcome. rate or number of businesses formally registered and tracked by government.", "level": 4, "type": "outcome"}}
+{"text": "regular cleaning", "id": 73, "metadata": {"definition": "outcome. measures of regular cleaning and maintenance behaviour.", "level": 4, "type": "outcome"}}
+{"text": "index of household wealth or assets", "id": 74, "metadata": {"definition": "outcome. composite/index measures of household saving and asset accumulation.", "level": 4, "type": "outcome"}}
+{"text": "index of confidence in public institutions", "id": 75, "metadata": {"definition": "outcome. composite/index measures of public confidence in public institutions", "level": 4, "type": "outcome"}}
+{"text": "employee training", "id": 76, "metadata": {"definition": "outcome. any measure of employee capacity building/training on the job.", "level": 4, "type": "outcome"}}
+{"text": "perception of corruption", "id": 77, "metadata": {"definition": "outcome. public perception of corruption levels of public servants and elected representatives.", "level": 4, "type": "outcome"}}
+{"text": "employment hours", "id": 78, "metadata": {"definition": "outcome. number of hours worked in specific time frame, usually either in a week or a month.", "level": 4, "type": "outcome"}}

models/retriever/level-4-small-no-negative-outcomes/document_index_db/embeddings.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1216e286a27fa8bb7445a17ab347c97cdabfde137b2a9538dff742fe84bfc508
+size 122539

models/retriever/level-4-small-no-negative-outcomes/question_encoder/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "intfloat/e5-small-v2",
+  "architectures": [
+    "GoldenRetrieverModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "auto_map": {
+    "AutoModel": "hf.GoldenRetrieverModel"
+  },
+  "classifier_dropout": null,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 384,
+  "initializer_range": 0.02,
+  "intermediate_size": 1536,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "projection_dim": null,
+  "torch_dtype": "float32",
+  "transformers_version": "4.33.3",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

models/retriever/level-4-small-no-negative-outcomes/question_encoder/hf.py

@@ -0,0 +1,99 @@
+from typing import Tuple, Union
+
+import torch
+from transformers import PretrainedConfig
+from transformers.modeling_outputs import BaseModelOutputWithPoolingAndCrossAttentions
+from transformers.models.bert.modeling_bert import BertModel
+
+
+class GoldenRetrieverConfig(PretrainedConfig):
+    model_type = "bert"
+
+    def __init__(
+        self,
+        vocab_size=30522,
+        hidden_size=768,
+        num_hidden_layers=12,
+        num_attention_heads=12,
+        intermediate_size=3072,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        max_position_embeddings=512,
+        type_vocab_size=2,
+        initializer_range=0.02,
+        layer_norm_eps=1e-12,
+        pad_token_id=0,
+        position_embedding_type="absolute",
+        use_cache=True,
+        classifier_dropout=None,
+        projection_dim=None,
+        **kwargs,
+    ):
+        super().__init__(pad_token_id=pad_token_id, **kwargs)
+
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.hidden_act = hidden_act
+        self.intermediate_size = intermediate_size
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.max_position_embeddings = max_position_embeddings
+        self.type_vocab_size = type_vocab_size
+        self.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.position_embedding_type = position_embedding_type
+        self.use_cache = use_cache
+        self.classifier_dropout = classifier_dropout
+        self.projection_dim = projection_dim
+
+
+class GoldenRetrieverModel(BertModel):
+    config_class = GoldenRetrieverConfig
+
+    def __init__(self, config, *args, **kwargs):
+        super().__init__(config)
+        self.layer_norm_layer = torch.nn.LayerNorm(
+            config.hidden_size, eps=config.layer_norm_eps
+        )
+        self.projection: torch.nn.Module | None = None
+        if config.projection_dim is not None:
+            self.projection = torch.nn.Sequential(
+                torch.nn.Linear(config.hidden_size, config.projection_dim),
+                torch.nn.LayerNorm(config.projection_dim),
+            )
+
+    def forward(
+        self, **kwargs
+    ) -> Union[Tuple[torch.Tensor], BaseModelOutputWithPoolingAndCrossAttentions]:
+        attention_mask = kwargs.get("attention_mask", None)
+        model_outputs = super().forward(**kwargs)
+        if attention_mask is None:
+            pooler_output = model_outputs.pooler_output
+        else:
+            token_embeddings = model_outputs.last_hidden_state
+            input_mask_expanded = (
+                attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
+            )
+            pooler_output = torch.sum(
+                token_embeddings * input_mask_expanded, 1
+            ) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
+
+        pooler_output = self.layer_norm_layer(pooler_output)
+
+        if self.projection is not None:
+            pooler_output = self.projection(pooler_output)
+
+        if not kwargs.get("return_dict", True):
+            return (model_outputs[0], pooler_output) + model_outputs[2:]
+
+        return BaseModelOutputWithPoolingAndCrossAttentions(
+            last_hidden_state=model_outputs.last_hidden_state,
+            pooler_output=pooler_output,
+            past_key_values=model_outputs.past_key_values,
+            hidden_states=model_outputs.hidden_states,
+            attentions=model_outputs.attentions,
+            cross_attentions=model_outputs.cross_attentions,
+        )

models/retriever/level-4-small-no-negative-outcomes/question_encoder/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2ff7f63276ef10722db26f66a406bc4aa6f5a345dd177d243cb826bb7db48b7
+size 133510894

models/retriever/level-4-small-no-negative-outcomes/question_encoder/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

models/retriever/level-4-small-no-negative-outcomes/question_encoder/tokenizer_config.json

@@ -0,0 +1,13 @@
+{
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}