Add message translation to french

DESCRIPTION

@@ -3,7 +3,8 @@ Title: Minimal Deep Learning Model Implementations
 Version: 0.0.0.9000
 Authors@R: c(
     person("Daniel", "Falbel", , "[email protected]", role = c("aut", "cre")),
-    person(family = "Posit", role = c("cph"))
+    person(family = "Posit", role = c("cph")),
+    person("Christophe", "Regouby", role = c("ctb"))
     )
 Description: A collection of minimal implementations of deep learning
     models. Clean and readable code prioritizing simplicity and understanding 

NEWS.md

@@ -0,0 +1,4 @@
+# minhub (development version)
+
+* Add support for `gpt2`, `gptneox`, `gptbigcode` and `llama`.
+* Add message translation to french (#3 @cregouby).

R/gpt2.R

@@ -197,21 +197,25 @@ gpt2_from_config <- function(identifier, revision = "main") {
   config <- jsonlite::fromJSON(path)
 
   if (config$model_type != "gpt2")
-    cli::cli_abort(c(
-      "{.arg config$model_type} must be {.val gpt2}, got {.val {config$model_type}}"
+    cli::cli_abort(gettext(
+      "{.arg config$model_type} must be {.val gpt2}, got {.val {config$model_type}}",
+      domain = "R-minhub"
     ))
 
   if (config$layer_norm_eps != 1e-5)
-    cli::cli_abort("{.arg config$layer_norm_eps} must be {.val 1e-5}.")
+    cli::cli_abort(gettext("{.arg config$layer_norm_eps} must be {.val 1e-5}.",
+                           domain = "R-minhub"))
 
   pdrop <- unlist(config[c("resid_pdrop", "embd_pdrop", "attn_pdrop")])
   if (length(unique(pdrop)) != 1)
-    cli::cli_abort("{.arg {names(pdrop)}} must be all equal, but got {pdrop}")
+    cli::cli_abort(gettext("{.arg {names(pdrop)}} must be all equal, but got {pdrop}",
+                           domain = "R-minhub"))
   else
     pdrop <- unique(pdrop)
 
   if (config$initializer_range != 0.02)
-    cli::cli_abort("{.arg initializer_range} must be {.val 0.02}, got {config$initializer_range}")
+    cli::cli_abort(gettext("{.arg initializer_range} must be {.val 0.02}, got {config$initializer_range}",
+                   domain = "R-minhub"))
 
   vocab_size <- config$vocab_size
   n_embd     <- config$n_embd

R/gptbigcode.R

@@ -135,27 +135,31 @@ gptbigcode_from_config <- function(identifier, revision = "main") {
   config <- jsonlite::fromJSON(path)
 
   if (config$model_type != "gpt_bigcode")
-    cli::cli_abort(c(
+    cli::cli_abort(gettext(
       x = "{.arg config$model_type} must be {.val gpt_bigcode}.",
-      i = "Got {.val {config$model_type}}"
+      i = "Got {.val {config$model_type}}",
+      domain = "R-minhub"
     ))
 
   if (!config$multi_query)
-    cli::cli_abort("Must use {.arg config$multi_query} but got {.val FALSE}")
+    cli::cli_abort(gettext("Must use {.arg config$multi_query} but got {.val FALSE}",
+                   domain = "R-minhub"))
 
   dropouts <- config[c("attn_pdrop", "resid_pdrop", "embd_pdrop")]
   if (length(unique(dropouts)) != 1)
-    cli::cli_abort(c(
+    cli::cli_abort(gettext(
       x = "All dropout must be equal.",
-      i = "Got {.val {names(dropouts)}} respectively {.val {dropouts}}"
+      i = "Got {.val {names(dropouts)}} respectively {.val {dropouts}}",
+      domain = "R-minhub"
     ))
   else
     pdrop <- unique(dropouts)
 
 
   if (config$layer_norm_eps != 1e-5)
-    cli::cli_abort(c(
-      x = "{.arg config$layer_norm_eps} must be 1e-5, got {.val {config$layer_norm_eps}}"
+    cli::cli_abort(gettext(
+      x = "{.arg config$layer_norm_eps} must be 1e-5, got {.val {config$layer_norm_eps}}",
+      domain = "R-minhub"
     ))
 
   # remap HF config attributes to minhub configurations

R/gptneox.R

@@ -93,7 +93,7 @@ nn_gptneox_attention <- nn_module(
     att <- torch_matmul(q, k$transpose(-2, -1)) * (1 / sqrt(k$size(-1)))
     att <- att$masked_fill(self$bias[,,1:t, 1:t] == 0, self$masked_bias)
     att <- nnf_softmax(att, dim=-1)$to(dtype = v$dtype)
-    
+
     y <- torch_matmul(att, v)$transpose(2, 3)$contiguous()$view(c(b, t, h))
     self$c_proj(y)
   }
@@ -178,32 +178,37 @@ gptneox_from_config <- function(identifier, revision = "main") {
   config <- jsonlite::fromJSON(path)
 
   if (config$model_type != "gpt_neox")
-    cli::cli_abort(c(
-      "{.arg config$model_type} must be {.val gpt_neox}, got {.val {config$model_type}}"
+    cli::cli_abort(gettext(
+      "{.arg config$model_type} must be {.val gpt_neox}, got {.val {config$model_type}}",
+      domain = "R-minhub"
     ))
 
   # parallel residual is not supported
   if (!config$use_parallel_residual)
-    cli::cli_abort(c(
+    cli::cli_abort(gettext(
       x = "Non parallel residual is not supported.",
-      i = "{.arg config$use_parallel_residual} is {.val FALSE}"
+      i = "{.arg config$use_parallel_residual} is {.val FALSE}",
+      domain = "R-minhub"
     ))
 
   if (config$hidden_act != "gelu")
-    cli::cli_abort(c(
+    cli::cli_abort(gettext(
       x = "Unsupported {.arg config$hidden_act}: {.val {config$hidden_act}}",
-      i = "Currently only {.val gelu} is supported."
+      i = "Currently only {.val gelu} is supported.",
+      domain = "R-minhub"
     ))
 
   if ((config$intermediate_size / config$hidden_size) != 4)
-    cli::cli_abort(c(
+    cli::cli_abort(gettext(
       x = "{.arg config$intermediate_size} must be 4*{.arg config$hidden_size}",
-      i = "Got {.val {config$intermediate_size}} and {.val {config$hidden_size}}"
+      i = "Got {.val {config$intermediate_size}} and {.val {config$hidden_size}}",
+      domain = "R-minhub"
     ))
 
   if (config$layer_norm_eps != 1e-5)
-    cli::cli_abort(c(
-      x = "{.arg config$layer_norm_eps} must be 1e-5, got {.val {config$layer_norm_eps}}"
+    cli::cli_abort(gettext(
+      x = "{.arg config$layer_norm_eps} must be 1e-5, got {.val {config$layer_norm_eps}}",
+      domain = "R-minhub"
     ))
 
   # remap HF config attributes to minhub configurations

R/llama.R

@@ -205,14 +205,16 @@ llama_from_config <- function(identifier, revision = "main") {
   config <- jsonlite::fromJSON(path)
 
   if (config$model_type != "llama")
-    cli::cli_abort(c(
-      "{.arg config$model_type} must be {.val llama}, got {.val {config$model_type}}"
+    cli::cli_abort(gettext(
+      "{.arg config$model_type} must be {.val llama}, got {.val {config$model_type}}",
+      domain = "R-minhub"
     ))
 
   if (config$hidden_act != "silu")
-    cli::cli_abort(c(
+    cli::cli_abort(gettext(
       x = "Unsupported {.arg config$hidden_act}: {.val {config$hidden_act}}",
-      i = "Currently only {.val silu} is supported."
+      i = "Currently only {.val silu} is supported.",
+      domain = "R-minhub"
     ))
 
   # remap HF config attributes to minhub configurations

R/weights.R

@@ -29,9 +29,10 @@ hf_state_dict <- function(identifier, revision = "main") {
     index_path <- tryCatch({
       hub_download(identifier, WEIGHTS_INDEX_NAME(), revision = revision)
     }, error = function(e) {
-      cli::cli_abort(c(
+      cli::cli_abort(gettext(
         x = "Error downloading weights from {.val {c(WEIGHTS_NAME(), WEIGHTS_INDEX_NAME())}}",
-        i = "Traceback below shows the error when trying to download {.val {WEIGHTS_NAME()}}"
+        i = "Traceback below shows the error when trying to download {.val {WEIGHTS_NAME()}}",
+        domain = "R-minhub"
       ), parent = err)
     })
 
@@ -70,7 +71,9 @@ state_dict_safetensors <- function(identifier, revision) {
   )
 
   if (inherits(index_path, "try-error")) {
-    cli::cli_abort("No safetensors files found.")
+    cli::cli_abort(gettext(
+      "No safetensors files found.",
+      domain = "R-minhub"))
   }
 
   index <- jsonlite::fromJSON(index_path)$weight_map %>%

po/R-fr.po

@@ -0,0 +1,129 @@
+msgid ""
+msgstr ""
+"Project-Id-Version: minhub 0.0.0.9000\n"
+"POT-Creation-Date: 2024-11-28 11:28+0100\n"
+"PO-Revision-Date: 2024-12-02 17:11+0100\n"
+"Last-Translator: \n"
+"Language-Team: \n"
+"Language: fr\n"
+"MIME-Version: 1.0\n"
+"Content-Type: text/plain; charset=UTF-8\n"
+"Content-Transfer-Encoding: 8bit\n"
+"X-Generator: Poedit 3.5\n"
+
+#: gpt2.R:201
+msgid ""
+"{.arg config$model_type} must be {.val gpt2}, got {.val {config$model_type}}"
+msgstr ""
+"La valeur de {.arg config$model_type} doit être {.val gpt2}, or elle est à {."
+"val {config$model_type}}"
+
+#: gpt2.R:205
+msgid "{.arg config$layer_norm_eps} must be {.val 1e-5}."
+msgstr "{.arg config$layer_norm_eps} doit être à {.val 1e-5}."
+
+#: gpt2.R:209
+msgid "{.arg {names(pdrop)}} must be all equal, but got {pdrop}"
+msgstr "Les {.arg {names(pdrop)}} doivent être égaux, or ils sont à {pdrop}"
+
+#: gpt2.R:214
+msgid ""
+"{.arg initializer_range} must be {.val 0.02}, got {config$initializer_range}"
+msgstr ""
+"La valeur de {.arg initializer_range} doit être à {.val 0.02}, or elle est à "
+"{config$initializer_range}"
+
+#: gptbigcode.R:139
+msgid "{.arg config$model_type} must be {.val gpt_bigcode}."
+msgstr "{.arg config$model_type} doit être {.val gpt_bigcode}."
+
+#: gptbigcode.R:140
+msgid "Got {.val {config$model_type}}"
+msgstr "Or elle est à  {.val {config$model_type}}"
+
+#: gptbigcode.R:144
+msgid "Must use {.arg config$multi_query} but got {.val FALSE}"
+msgstr ""
+"Il faut utiliser {.arg config$multi_query}, or c'est actuellement {.val "
+"FALSE}"
+
+#: gptbigcode.R:149
+msgid "All dropout must be equal."
+msgstr "Tous les {.val dropout} doivent être égaux."
+
+#: gptbigcode.R:150
+msgid "Got {.val {names(dropouts)}} respectively {.val {dropouts}}"
+msgstr "Or {.val {names(dropouts)}} sont respectivement à  {.val {dropouts}}"
+
+#: gptbigcode.R:158 gptneox.R:206
+msgid ""
+"{.arg config$layer_norm_eps} must be 1e-5, got {.val {config$layer_norm_eps}}"
+msgstr ""
+"La valeur de {.arg config$layer_norm_eps} doit être {.val 1e-5}, or elle est "
+"à {.val {config$layer_norm_eps}}"
+
+#: gptneox.R:182
+msgid ""
+"{.arg config$model_type} must be {.val gpt_neox}, got {.val "
+"{config$model_type}}"
+msgstr ""
+"{.arg config$model_type} doit être {.val gpt_neox}, or elle est à {.val "
+"{config$model_type}}"
+
+#: gptneox.R:188
+msgid "Non parallel residual is not supported."
+msgstr "Les résidus non-parallèles ne sont pas pris en compte."
+
+#: gptneox.R:189
+msgid "{.arg config$use_parallel_residual} is {.val FALSE}"
+msgstr "{.arg config$use_parallel_residual} est {.val FALSE}"
+
+#: gptneox.R:194 llama.R:214
+msgid "Unsupported {.arg config$hidden_act}: {.val {config$hidden_act}}"
+msgstr "{.arg config$hidden_act} ne peut pas être {.val {config$hidden_act}}"
+
+#: gptneox.R:195
+msgid "Currently only {.val gelu} is supported."
+msgstr "Actuellement seul {.val gelu} est pris en compte."
+
+#: gptneox.R:200
+msgid "{.arg config$intermediate_size} must be 4*{.arg config$hidden_size}"
+msgstr ""
+"{.arg config$intermediate_size} doit être de  4*{.arg config$hidden_size}"
+
+#: gptneox.R:201
+msgid "Got {.val {config$intermediate_size}} and {.val {config$hidden_size}}"
+msgstr ""
+"Or nous avons respectivement {.val {config$intermediate_size}} et {.val "
+"{config$hidden_size}}"
+
+#: llama.R:209
+msgid ""
+"{.arg config$model_type} must be {.val llama}, got {.val {config$model_type}}"
+msgstr ""
+"{.arg config$model_type} doit être {.val llama}, or nous avons {.val "
+"{config$model_type}}"
+
+#: llama.R:215
+msgid "Currently only {.val silu} is supported."
+msgstr "Actuellement seul {.val silu} est pris en compte."
+
+#: weights.R:33
+msgid ""
+"Error downloading weights from {.val {c(WEIGHTS_NAME(), "
+"WEIGHTS_INDEX_NAME())}}"
+msgstr ""
+"Problème de téléchargement des points depuis {.val {c(WEIGHTS_NAME(), "
+"WEIGHTS_INDEX_NAME())}}"
+
+#: weights.R:34
+msgid ""
+"Traceback below shows the error when trying to download {.val "
+"{WEIGHTS_NAME()}}"
+msgstr ""
+"Le log suivant détaille l'erreur lors de la tentative de téléchargement de {."
+"val {WEIGHTS_NAME()}}"
+
+#: weights.R:73
+msgid "No safetensors files found."
+msgstr "Aucun fichier safetensors trouvé."