add pdf support for Claude

README.md

@@ -182,6 +182,38 @@ This generates the expected cache block below
 }
 ```
 
+#### PDF chats
+
+*msglm* offers PDF
+[support](https://docs.anthropic.com/en/docs/build-with-claude/pdf-support)
+for Anthropic. Just like an image chat all you need to do is pass the
+raw pdf bytes in a list with your question to *mk_msg* and it will
+generate the correct format as shown in the example below.
+
+``` python
+import httpx
+from msglm import mk_msg_anthropic as mk_msg
+from anthropic import Anthropic
+
+client = Anthropic(default_headers={'anthropic-beta': 'pdfs-2024-09-25'})
+
+url = "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-October-Addendum.pdf"
+pdf = httpx.get(url).content
+
+r = client.messages.create(
+    model="claude-3-5-sonnet-20241022",
+    max_tokens=1024,
+    messages=[mk_msg([pdf, "Which model has the highest human preference win rates across each use-case?"])]
+)
+print(r.content[0].text)
+```
+
+Note: this feature is currently in beta so you’ll need to:
+
+- use the Anthropic beta client
+  (e.g. `anthropic.Anthropic(default_headers={'anthropic-beta': 'pdfs-2024-09-25'})`)
+- use the `claude-3-5-sonnet-20241022` model
+
 ### Summary
 
 We hope *msglm* will make your life a little easier when chatting to

msglm/_modidx.py

@@ -9,19 +9,24 @@
                             'msglm.core.AnthropicMsg.find_block': ('core.html#anthropicmsg.find_block', 'msglm/core.py'),
                             'msglm.core.AnthropicMsg.img_msg': ('core.html#anthropicmsg.img_msg', 'msglm/core.py'),
                             'msglm.core.AnthropicMsg.is_sdk_obj': ('core.html#anthropicmsg.is_sdk_obj', 'msglm/core.py'),
+                            'msglm.core.AnthropicMsg.pdf_msg': ('core.html#anthropicmsg.pdf_msg', 'msglm/core.py'),
                             'msglm.core.Msg': ('core.html#msg', 'msglm/core.py'),
                             'msglm.core.Msg.__call__': ('core.html#msg.__call__', 'msglm/core.py'),
                             'msglm.core.Msg.find_block': ('core.html#msg.find_block', 'msglm/core.py'),
                             'msglm.core.Msg.img_msg': ('core.html#msg.img_msg', 'msglm/core.py'),
                             'msglm.core.Msg.is_sdk_obj': ('core.html#msg.is_sdk_obj', 'msglm/core.py'),
                             'msglm.core.Msg.mk_content': ('core.html#msg.mk_content', 'msglm/core.py'),
+                            'msglm.core.Msg.pdf_msg': ('core.html#msg.pdf_msg', 'msglm/core.py'),
                             'msglm.core.Msg.text_msg': ('core.html#msg.text_msg', 'msglm/core.py'),
                             'msglm.core.OpenAiMsg': ('core.html#openaimsg', 'msglm/core.py'),
                             'msglm.core.OpenAiMsg.find_block': ('core.html#openaimsg.find_block', 'msglm/core.py'),
                             'msglm.core.OpenAiMsg.img_msg': ('core.html#openaimsg.img_msg', 'msglm/core.py'),
                             'msglm.core.OpenAiMsg.is_sdk_obj': ('core.html#openaimsg.is_sdk_obj', 'msglm/core.py'),
                             'msglm.core._add_cache_control': ('core.html#_add_cache_control', 'msglm/core.py'),
+                            'msglm.core._is_img': ('core.html#_is_img', 'msglm/core.py'),
+                            'msglm.core._is_pdf': ('core.html#_is_pdf', 'msglm/core.py'),
                             'msglm.core._mk_img': ('core.html#_mk_img', 'msglm/core.py'),
+                            'msglm.core._mk_pdf': ('core.html#_mk_pdf', 'msglm/core.py'),
                             'msglm.core.mk_msg': ('core.html#mk_msg', 'msglm/core.py'),
                             'msglm.core.mk_msg_anthropic': ('core.html#mk_msg_anthropic', 'msglm/core.py'),
                             'msglm.core.mk_msgs': ('core.html#mk_msgs', 'msglm/core.py'),

msglm/core.py

@@ -22,6 +22,17 @@ def _mk_img(data:bytes)->tuple:
     mtype = mimetypes.types_map["."+imghdr.what(None, h=data)]
     return img, mtype
 
+# %% ../nbs/00_core.ipynb
+def _is_img(data): return isinstance(data, bytes) and bool(imghdr.what(None, data))
+
+# %% ../nbs/00_core.ipynb
+def _is_pdf(data): return isinstance(data, bytes) and data.startswith(b'%PDF-')
+
+# %% ../nbs/00_core.ipynb
+def _mk_pdf(data:bytes)->str:
+    "Convert pdf bytes to a base64 encoded pdf"
+    return base64.standard_b64encode(data).decode("utf-8")
+
 # %% ../nbs/00_core.ipynb
 def mk_msg(content:Union[list,str], role:str="user", *args, api:str="openai", **kw)->dict:
     "Create an OpenAI/Anthropic compatible message."
@@ -64,10 +75,15 @@ def img_msg(self, *args, **kw)->dict:
         "Convert bytes to an image message"
         raise NotImplemented
 
+    def pdf_msg(self, *args, **kw)->dict:
+        "Convert bytes to a pdf message"
+        raise NotImplemented
+
     def mk_content(self, content:[str, bytes], text_only:bool=False) -> dict:
         "Create the appropriate data structure based the content type."
         if isinstance(content, str): return self.text_msg(content, text_only=text_only)
-        if isinstance(content, bytes): return self.img_msg(content)
+        if _is_img(content): return self.img_msg(content)
+        if _is_pdf(content): return self.pdf_msg(content)
         return content
 
 # %% ../nbs/00_core.ipynb
@@ -79,6 +95,11 @@ def img_msg(self, data: bytes) -> dict:
         r = {"type": "base64", "media_type": mtype, "data":img}
         return {"type": "image", "source": r}
 
+    def pdf_msg(self, data: bytes) -> dict:
+        "Convert `data` to a pdf message"
+        r = {"type": "base64", "media_type": "application/pdf", "data":_mk_pdf(data)}
+        return {"type": "document", "source": r}
+
     def is_sdk_obj(self, r)-> bool:
         "Check if `r` is an SDK object."
         return isinstance(r, abc.Mapping)