-
Notifications
You must be signed in to change notification settings - Fork 654
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
[Proposal] Hybrid architectures warn on unresolved optional 'hook_attn' aliases when they are skipped
bugSomething isn't workingSomething isn't workingcomplexity-simpleSimple issues, which may be good for beginnersSimple issues, which may be good for beginnersTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.#1578 In TransformerLensOrg/TransformerLens;[Proposal] Native adapter: param-free pre-norm (LNPre/RMSPre) + fold_ln support
complexity-highVery complicated changes for people to address who are quite familiar with the codeVery complicated changes for people to address who are quite familiar with the codeenhancementNew feature or requestNew feature or requesthelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Bug Report] Native init drifts from
HookedTransformer: unresolvedinitializer_rangesentinel and ignored xavier/kaiming gaincomplexity-moderateModerately complicated issues for people who have intermediate experience with the codeModerately complicated issues for people who have intermediate experience with the codegood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Proposal] Honor cfg.init_weights=False in boot_native and add bridge.init_weights()
complexity-moderateModerately complicated issues for people who have intermediate experience with the codeModerately complicated issues for people who have intermediate experience with the codeenhancementNew feature or requestNew feature or requestgood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Proposal] Rewrite executable doctests that construct HookedTransformer onto TransformerBridge
complexity-moderateModerately complicated issues for people who have intermediate experience with the codeModerately complicated issues for people who have intermediate experience with the codeenhancementNew feature or requestNew feature or requestgood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Proposal] Regression tests for Hooked* deprecation warnings + scoped filterwarnings policy
complexity-simpleSimple issues, which may be good for beginnersSimple issues, which may be good for beginnersgood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Bug Report] boot_native fails with an opaque AttributeError on legacy config objects
bugSomething isn't workingSomething isn't workingcomplexity-simpleSimple issues, which may be good for beginnersSimple issues, which may be good for beginnersgood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Proposal] Add a runnable W_pos / W_E_pos migration snippet to the migration guide
complexity-simpleSimple issues, which may be good for beginnersSimple issues, which may be good for beginnersdocumentationImprovements or additions to documentationImprovements or additions to documentationgood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Proposal] Migration recipes for the new bridge capabilities in migrating_to_v3.md
complexity-simpleSimple issues, which may be good for beginnersSimple issues, which may be good for beginnersdocumentationImprovements or additions to documentationImprovements or additions to documentationgood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Bug Report] Bridge: start_at_layer on audio models crashes inside HF Conv1d
bugSomething isn't workingSomething isn't workingcomplexity-moderateModerately complicated issues for people who have intermediate experience with the codeModerately complicated issues for people who have intermediate experience with the codegood first issueGood for newcomersGood for newcomershelp wantedExtra attention is neededExtra attention is neededhigh-priorityMaintainers are interested in these issues being solved before othersMaintainers are interested in these issues being solved before othersTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Bug Report] Bridge W_in/W_out returned in torch orientation on nn.Linear-backed models
bugSomething isn't workingSomething isn't workingcomplexity-moderateModerately complicated issues for people who have intermediate experience with the codeModerately complicated issues for people who have intermediate experience with the codehelp wantedExtra attention is neededExtra attention is neededhigh-priorityMaintainers are interested in these issues being solved before othersMaintainers are interested in these issues being solved before othersTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.[Bug Report] Bridge QK/OV weight circuits broken on GQA models
bugSomething isn't workingSomething isn't workinghelp wantedExtra attention is neededExtra attention is neededhigh-priorityMaintainers are interested in these issues being solved before othersMaintainers are interested in these issues being solved before othersTransformerBridgeBug specific to the new TransformerBridge systemBug specific to the new TransformerBridge systemStatus: Open.