Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74696.70782709865,
-      "required_pages_per_second_upper_estimate": 0.006474521317377744
+      "required_token_positions_per_second": 74672.59742489789,
+      "required_pages_per_second_upper_estimate": 0.006472431488822238
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9337.088478387332,
-      "required_pages_per_second_upper_estimate": 0.0037247041959419707
+      "required_token_positions_per_second": 9334.074678112236,
+      "required_pages_per_second_upper_estimate": 0.003723501945951905
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18674.176956774663,
-      "required_pages_per_second_upper_estimate": 0.004319942184660317
+      "required_token_positions_per_second": 18668.149356224472,
+      "required_pages_per_second_upper_estimate": 0.0043185478053551765
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28011.265435161993,
-      "required_pages_per_second_upper_estimate": 0.0017057391012495842
+      "required_token_positions_per_second": 28002.224034336712,
+      "required_pages_per_second_upper_estimate": 0.0017051885273758887
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28011.265435161993,
-      "required_pages_per_second_upper_estimate": 0.002156384753344285
+      "required_token_positions_per_second": 28002.224034336712,
+      "required_pages_per_second_upper_estimate": 0.0021556887212805555
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28011.265435161993,
-      "required_pages_per_second_upper_estimate": 0.3478061690299116
+      "required_token_positions_per_second": 28002.224034336712,
+      "required_pages_per_second_upper_estimate": 0.3476939050912837
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3062706661561947,
-    "producer_p95_seconds": 0.22277083799963293,
-    "gpu_update_seconds": 0.9870314521847344,
-    "producer_batches_per_second": 3.265085790129215,
-    "gpu_batches_per_second": 1.0131389407972364,
-    "producer_headroom": 3.22274236893898,
+    "producer_batch_seconds": 0.2805528774842969,
+    "producer_p95_seconds": 0.21641377599917178,
+    "gpu_update_seconds": 0.9873501464061442,
+    "producer_batches_per_second": 3.5643904598910128,
+    "gpu_batches_per_second": 1.0128119225382202,
+    "producer_headroom": 3.519301442422055,
     "sustainable": true,
-    "buffer_seconds": 60.2089185832688,
+    "buffer_seconds": 60.228358930774796,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 227954.08391677495,
-    "upload_bytes_per_second": 4969578.80704246,
+    "generation_bytes_per_second": 259982.28901204382,
+    "upload_bytes_per_second": 4956588.409946015,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.91363990905224,
+    "estimated_full_shards_per_hour": 55.78614714404093,
     "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": 0.9417934939999668,
-      "bytes": 2584042,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 0.966999364000003,
       "bytes": 6885855,
@@ -562,6 +556,12 @@
       "bytes": 7577178,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0859860309992655,
+      "bytes": 2128705,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -79,3 +79,4 @@
 {"step": 7900, "loss": 0.7169801648706198, "lr": 0.0003, "grad_norm": 0.1563463019300498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1155501400, "tokens_per_second": 140286.4032580151, "data_wait_s": 1.5573999917251058e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 8000, "loss": 0.7449607010930777, "lr": 0.0003, "grad_norm": 0.14968626283655082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1170105212, "tokens_per_second": 148722.72278340437, "data_wait_s": 1.4211000234354287e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 8100, "loss": 0.693482581526041, "lr": 0.0003, "grad_norm": 0.1611827912346656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1184819071, "tokens_per_second": 149585.46369521818, "data_wait_s": 9.639001291361637e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 8200, "loss": 0.7218140847980976, "lr": 0.0003, "grad_norm": 0.15468966826019376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1199413328, "tokens_per_second": 148992.5304126804, "data_wait_s": 7.2950006142491475e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7cc4ccd4f112abdaf8b9b40a542fce1812757448c89b31d31c6a7dd935e9180
-size 169622
+oid sha256:43f4baaa40f2bf260f41c3541419ba60b0bceb432e19687f12cbeba4d55eb2d7
+size 171705