From 546ad43b152bbd9bab5367d79c47bf37e10c33f7 Mon Sep 17 00:00:00 2001
From: Krrish Dholakia <krrishdholakia@gmail.com>
Date: Tue, 29 Aug 2023 16:46:04 -0700
Subject: [PATCH] adding context window exceeded error to huggingface

---
 litellm/__pycache__/utils.cpython-311.pyc | Bin 75377 -> 75648 bytes
 litellm/llms/huggingface_restapi.py       |   5 ++++-
 litellm/utils.py                          |   7 +++++++
 3 files changed, 11 insertions(+), 1 deletion(-)

diff --git a/litellm/__pycache__/utils.cpython-311.pyc b/litellm/__pycache__/utils.cpython-311.pyc
index 14fa58e19d5c71d13fba1986367a7b354ba19400..36517095ba7e35980ddf5be1559d7713fb7e6f5b 100644
GIT binary patch
delta 1073
zcmZ`&Sxi$w6rDTowPmX;T}ZJ)q0~~a0#Y_j6pY3gf^mz%#19BlvC0zLy1~;BK8Swk
z&uAo3L;OLFXe9H|L}SDcOhmz!EsaP-#b{jHxHK*iryy>?&Ajto-kfvK+<P;@`(pD0
zalr+pG79nd(ejRLE$v!xv=?uUVQsxDKnS^&v}A1(3*0Nm&`nqNmqI&h?$>}I)T?OQ
zgJ+A*OPCn<;IQ6#$88C}r-P(RBHi1gP9`Qv+>e}Uf)5Ms_?c~lBvC@dNRtLu$rekn
zJC+7}lBqOc0yFh`%Q)(N<x}73^n_25nDNWx=|3hG8vjL2yM0E;q!YeWO%~6jobfn=
z&SCD6G225mi^o|deJQYnR`%)U$&%wMo>RABFcGYD=U{I2Qc2D$HIr;fPL}Fsj)C~U
zMgLb%a{lG~I;*u))*&EtnLpFf0%SVtVv-wH(656j`EzKAwKJiEkKyvvD8ky`N;6qD
zrQ9%-QaXo{cW5rxys26&T$%T$5@MrQhjfYaWTD{><@`SZV&$Dx)0M-8iu|)!Qgjwj
zX?P11QtxmX6jA-iCMah0BRX6VYj-*uoQ68P%T-llUv6mB)Y|K7+`A05jyi|iU_V%G
zw-Y;QRJz<%?tLypbpzon+UnoTUiee7@k=B^&1gI&-7D$?xJi(i#-cS|ZN?R=FS=+b
zx+s8ZXx8X$*u>tB>cw?&d@t`}BnQ5Hakim2n>WtpS7lywcvXdb`Lpx9>Lp`IHow8*
zPd55Z%K{2<p*Dy(d<;VXi5e~4@?<@;jXTA}Z3)}7Xha)TVHQRVt8qn|;7^lPnAyz<
z1wbzAdCo8_XNO)M5MUz<zAnJvU>WbSF-Wu`a1&aoHrNBl*rVVez;Sl%LmGyY^kGQP
z-hX-p{Ng<fMPT_QIYO>#`Z%Pe4U@4<`E@%$7pwV}1K?q8-)~^W^)MqA=y*uO8YbV!
z;jU+g9AX4$@;D;l7KYCrOAMTn;RuDNJ098v$n;cdz#&4er&kAeqg6))Fa^+GxF+Is
IhY9Ze1Z?s<3;+NC

delta 887
zcmZ{iZAg<*6vv-)pS_g!ICM_g8qK$D>U>+J>jm=#A|d=>A<`FH_*9gTuuY4Yx}rev
zOK`FYne|BpMX-zQLxlQ4B2ArpF~mMpQe^ouC1R9W&!$$?Ja8}f-1GmPd-!u5Q=-+E
zqL^-ui(-U$-n`5^iw0s^#)v1HwvQbJ_KA(d;^QhB7%wN#jK?NRpqe@+6~K}!9&DeQ
z=gfA(r`D1XZ@G>A9t~J=-J^#zT%1nBGB1Z`Csab#Ib9yUGbC!a)bDH=A!;*t9VUH?
z$5Kx!=<$LlS&?ps%iMPLKJ$Rs8D5JWo+5FEP?6#4=sw)=XyyOu1Xb~b1~Sn+q1>U1
z*?%}9=k_Lo0Y7?;T$WwZz*zKrwq3$!q0Xl<<%CYk?SYuBi41+>E|y_>g|!(sf!}0x
zd~TSv`!x9xSl-YV*Cn{t8+I<l{wDQ&UYK<IG$j#8!G}n%8p9K32&D1Pl>C0o^D7f~
z=z<|1a{l$1Hwga4<FMZ(J<vhy(!PVZ;Xe)ecxvV(6kz8}1(;}HMoF-Ewj$;_XSJQP
z)mxKl>iF|V=$+XVBD=erVI0gGO`2Pz4iF2D%*mk|{c|HwMxC$KqByzAy*In3aavyJ
zlNSaUd8~?NzHJq$Dq|}Zj9(E|rDOe?Qc{)7zDgDgNe%0~0T7S#?*oudeIF<RBW+o@
z%EF1qO9ceX)U=XIgpYS|^%*pxc5M_K^z)h*;1(VFl0u*r7lLZK@%0nHJ-QIwMI?_T
z3}daq<zPZv6UeCg`x$^9+OVDn&`-O6JSEbBkU|Ey5>yN}!J-6)UA0o!!-B<iEeeJR
uEV&MG&?Y87C9bJBc*sJst1ba9i6GrItAZEN@_H7u0p_pWB6SYxVEh+U$_?ZI

diff --git a/litellm/llms/huggingface_restapi.py b/litellm/llms/huggingface_restapi.py
index fbde78334..385e3a810 100644
--- a/litellm/llms/huggingface_restapi.py
+++ b/litellm/llms/huggingface_restapi.py
@@ -94,7 +94,10 @@ class HuggingfaceRestAPILLM:
                 additional_args={"complete_input_dict": data},
             )
             ## RESPONSE OBJECT
-            completion_response = response.json()
+            try:
+                completion_response = response.json()
+            except:
+                raise HuggingfaceError(message=response.text, status_code=response.status_code)
             print_verbose(f"response: {completion_response}")
             if isinstance(completion_response, dict) and "error" in completion_response:
                 print_verbose(f"completion error: {completion_response['error']}")
diff --git a/litellm/utils.py b/litellm/utils.py
index ea54943fc..5fc175528 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -1462,6 +1462,13 @@ def exception_type(model, original_exception, custom_llm_provider):
                         llm_provider="cohere",
                     )
             elif custom_llm_provider == "huggingface":
+                if "length limit exceeded" in error_str:
+                    exception_mapping_worked = True
+                    raise ContextWindowExceededError(
+                        message=error_str,
+                        model=model,
+                        llm_provider="huggingface"
+                    )
                 if hasattr(original_exception, "status_code"):
                     if original_exception.status_code == 401:
                         exception_mapping_worked = True