Update README.md
Browse files
README.md
CHANGED
|
@@ -22,7 +22,7 @@ Lyra-v2a2
|
|
| 22 |
v
|
| 23 |
Lyra-v3
|
| 24 |
|
|
| 25 |
-
| [Backmerge + LoRA Extraction + Low Rank SFT Step]
|
| 26 |
v
|
| 27 |
Lyra-v4
|
| 28 |
```
|
|
@@ -64,9 +64,8 @@ min_p: 0.1 - 0.2 # Crucial for NeMo
|
|
| 64 |
# Notes
|
| 65 |
|
| 66 |
\- Some people have been having issues with run-on generations for Lyra-v3. Kind of weird, when I never had issues.
|
| 67 |
-
<br>\- Anyway, make sure not to skip special tokens, or ban EOS tokens. I think this is the main problem that happens when v3 was to be quanted. The special tokens map is fucked in v3, Quantizing tools likely spazzed out seeing it. I blame llamafactory for it. It ran fine unquantised.
|
| 68 |
<br>\- I like long generations, though I can control it easily to create short ones. If you're struggling, prompt better. Fix your system prompts, use an Author's Note, use a prefill. They are there for a reason.
|
| 69 |
-
<br>\- Lyra passes my internal benchmark suite, hence why I'm releasing it. Do I like it? Yes? it's out. that's it. They are my models for my enjoyment.
|
| 70 |
-
<br>\- Issues like roleplay format are what I consider worthless, as it follows few-shot examples fine. This is not a priority for me to 'fix', as I see no isses with it.
|
| 71 |
-
<br>\- If you don't like it, just try another model? Plenty of other choices. Ymmv, I like it.
|
| 72 |
-
<br>\- This may sound rough, but Roleplay is subjective. What I like, you may not like. It's fine.
|
|
|
|
| 22 |
v
|
| 23 |
Lyra-v3
|
| 24 |
|
|
| 25 |
+
| [Backmerge to v2a1 + LoRA Extraction + Low Rank SFT Step for Coherency]
|
| 26 |
v
|
| 27 |
Lyra-v4
|
| 28 |
```
|
|
|
|
| 64 |
# Notes
|
| 65 |
|
| 66 |
\- Some people have been having issues with run-on generations for Lyra-v3. Kind of weird, when I never had issues.
|
| 67 |
+
<br>\- Anyway, make sure not to skip special tokens, or ban EOS tokens. I think this is the main problem that happens when v3 was to be quanted. The special tokens map config is fucked in v3, Quantizing tools likely spazzed out seeing it. I blame llamafactory for it. It ran fine unquantised.
|
| 68 |
<br>\- I like long generations, though I can control it easily to create short ones. If you're struggling, prompt better. Fix your system prompts, use an Author's Note, use a prefill. They are there for a reason.
|
| 69 |
+
<br>\- Lyra passes my internal benchmark suite, hence why I'm releasing it. Do I like it? Yes? it's out. that's it. They are my models for my personal enjoyment first.
|
| 70 |
+
<br>\- Issues like roleplay format are what I consider worthless, as it follows few-shot examples fine. This is not a priority for me to 'fix', as I see no isses with it. Same with excessive generations. Its easy to cut out.
|
| 71 |
+
<br>\- If you don't like it, just try another model? Plenty of other choices. Ymmv, I like it.
|
|
|