TernaryPilot v2 adapter (rank 64, F16) and fixed benchmark runner
Browse files
.gitattributes
CHANGED
|
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
TernaryPilot-v2-Ternary-Bonsai-2-9B-LoRA-F16.gguf filter=lfs diff=lfs merge=lfs -text
|
TernaryPilot-v2-Ternary-Bonsai-2-9B-LoRA-F16.gguf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fedb5f7d7b3803165456558873d37643ce24efaefb0a263ca428092de7adad0b
|
| 3 |
+
size 320891744
|
training/banco_pubblico.py
CHANGED
|
@@ -121,9 +121,9 @@ class Server:
|
|
| 121 |
|
| 122 |
def __enter__(self):
|
| 123 |
a = self.a
|
| 124 |
-
cmd = [a.server, "-m", a.model, "--mmproj", a.mmproj, "--lora-scaled",
|
| 125 |
"--image-max-tokens", "1024", "-ngl", "999", "-fa", "on", "--jinja", "-c", str(16384 * a.parallel),
|
| 126 |
-
"--parallel", str(a.parallel), "--port", str(a.port)]
|
| 127 |
self.log = open(a.out + ".server.log", "w")
|
| 128 |
self.p = subprocess.Popen(cmd, stdout=self.log, stderr=subprocess.STDOUT)
|
| 129 |
for _ in range(900):
|
|
|
|
| 121 |
|
| 122 |
def __enter__(self):
|
| 123 |
a = self.a
|
| 124 |
+
cmd = [a.server, "-m", a.model, "--mmproj", a.mmproj, "--lora-scaled", ",".join(f"{x}:0" for x in a.lora),
|
| 125 |
"--image-max-tokens", "1024", "-ngl", "999", "-fa", "on", "--jinja", "-c", str(16384 * a.parallel),
|
| 126 |
+
"--parallel", str(a.parallel), "--port", str(a.port), "--cache-ram", "0"] # prompt cache in host RAM: 8 GiB per server by default
|
| 127 |
self.log = open(a.out + ".server.log", "w")
|
| 128 |
self.p = subprocess.Popen(cmd, stdout=self.log, stderr=subprocess.STDOUT)
|
| 129 |
for _ in range(900):
|