Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76845.03788351064,
-      "required_pages_per_second_upper_estimate": 0.005316318025834798
+      "required_token_positions_per_second": 76576.94871656822,
+      "required_pages_per_second_upper_estimate": 0.005297770995213153
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9605.62973543883,
-      "required_pages_per_second_upper_estimate": 0.004021676581869699
+      "required_token_positions_per_second": 9572.118589571028,
+      "required_pages_per_second_upper_estimate": 0.004007646164887153
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19211.25947087766,
-      "required_pages_per_second_upper_estimate": 0.004001198712801843
+      "required_token_positions_per_second": 19144.237179142056,
+      "required_pages_per_second_upper_estimate": 0.003987239736930008
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28816.88920631649,
-      "required_pages_per_second_upper_estimate": 0.001727528507076587
+      "required_token_positions_per_second": 28716.35576871308,
+      "required_pages_per_second_upper_estimate": 0.0017215016809979335
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28816.88920631649,
-      "required_pages_per_second_upper_estimate": 0.0022403863836910633
+      "required_token_positions_per_second": 28716.35576871308,
+      "required_pages_per_second_upper_estimate": 0.0022325703511172574
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28816.88920631649,
-      "required_pages_per_second_upper_estimate": 0.3696321135736649
+      "required_token_positions_per_second": 28716.35576871308,
+      "required_pages_per_second_upper_estimate": 0.3683425785804836
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11051932418746446,
-    "producer_p95_seconds": 0.3485527130033006,
-    "gpu_update_seconds": 0.9594373564076348,
-    "producer_batches_per_second": 9.04819141224376,
-    "gpu_batches_per_second": 1.0422775320571647,
-    "producer_headroom": 8.681172848833416,
+    "producer_batch_seconds": 0.27182619363310323,
+    "producer_p95_seconds": 0.36465414299891563,
+    "gpu_update_seconds": 0.9627962622653856,
+    "producer_batches_per_second": 3.6788213329792185,
+    "gpu_batches_per_second": 1.0386413400142174,
+    "producer_headroom": 3.541955428934555,
     "sustainable": true,
-    "buffer_seconds": 60.444553453680996,
+    "buffer_seconds": 60.656164522719294,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 572375.6149939366,
-    "upload_bytes_per_second": 5685007.255664664,
+    "generation_bytes_per_second": 555525.874211112,
+    "upload_bytes_per_second": 5766829.2342223395,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.40924137765681,
+    "estimated_full_shards_per_hour": 119.20292062521001,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.4250826019997476,
-      "bytes": 6016198,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.177016456000274,
       "bytes": 5382045,
@@ -562,6 +556,12 @@
       "bytes": 1233565,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.3546634440062917,
+      "bytes": 11809923,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -412,3 +412,4 @@
 {"step": 598300, "loss": 0.5658122301101685, "lr": 0.0003, "grad_norm": 0.16386031798808692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87573273734, "tokens_per_second": 152925.5204781657, "data_wait_s": 8.282993803732097e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 598400, "loss": 0.599250316619873, "lr": 0.0003, "grad_norm": 0.1595777341662248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87587958515, "tokens_per_second": 152418.13595460265, "data_wait_s": 1.4655000995844603e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 598500, "loss": 0.5546641945838928, "lr": 0.0003, "grad_norm": 0.1530256080927078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87602606207, "tokens_per_second": 151709.41766320533, "data_wait_s": 1.795100251911208e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 598600, "loss": 0.5612573027610779, "lr": 0.0003, "grad_norm": 0.14967737767813183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87617321719, "tokens_per_second": 152887.0579783995, "data_wait_s": 1.1379001080058515e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5cc8127063313dcae3352e1db4492a3b3489313b01535cf40105a0365c2b2a08
-size 883281
+oid sha256:d763b9422af8451355a2b16bd8dc3f04796837277933683ba96f3bc116fb7b89
+size 885399