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": 77063.94010796421,
-      "required_pages_per_second_upper_estimate": 0.0053314621896456355
+      "required_token_positions_per_second": 77324.04105181195,
+      "required_pages_per_second_upper_estimate": 0.005349456576458359
     },
     {
       "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": 9632.992513495527,
-      "required_pages_per_second_upper_estimate": 0.004033132805642256
+      "required_token_positions_per_second": 9665.505131476493,
+      "required_pages_per_second_upper_estimate": 0.004046745159850219
     },
     {
       "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": 19265.985026991053,
-      "required_pages_per_second_upper_estimate": 0.004012596602930296
+      "required_token_positions_per_second": 19331.010262952987,
+      "required_pages_per_second_upper_estimate": 0.004026139644750377
     },
     {
       "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": 28898.97754048658,
-      "required_pages_per_second_upper_estimate": 0.001732449577368455
+      "required_token_positions_per_second": 28996.51539442948,
+      "required_pages_per_second_upper_estimate": 0.0017382968227806523
     },
     {
       "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": 28898.97754048658,
-      "required_pages_per_second_upper_estimate": 0.002246768390604365
+      "required_token_positions_per_second": 28996.51539442948,
+      "required_pages_per_second_upper_estimate": 0.002254351529724746
     },
     {
       "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": 28898.97754048658,
-      "required_pages_per_second_upper_estimate": 0.37068505458481266
+      "required_token_positions_per_second": 28996.51539442948,
+      "required_pages_per_second_upper_estimate": 0.3719361654472041
     }
   ],
   "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": 0.1298623691875207,
-    "producer_p95_seconds": 0.2888023829982558,
-    "gpu_update_seconds": 0.9567120484199139,
-    "producer_batches_per_second": 7.700460158369699,
-    "gpu_batches_per_second": 1.0452465834956084,
-    "producer_headroom": 7.3671230118898094,
+    "producer_batch_seconds": 0.43811565751559556,
+    "producer_p95_seconds": 0.31185393199848477,
+    "gpu_update_seconds": 0.9534938810375627,
+    "producer_batches_per_second": 2.282502309254727,
+    "gpu_batches_per_second": 1.0487744283286127,
+    "producer_headroom": 2.176351985328489,
     "sustainable": true,
-    "buffer_seconds": 60.27285905045457,
+    "buffer_seconds": 60.070114505366455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352086.87916442344,
-    "upload_bytes_per_second": 6139443.606591037,
+    "generation_bytes_per_second": 322271.09524717537,
+    "upload_bytes_per_second": 6127140.705944653,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.54964810561682,
+    "estimated_full_shards_per_hour": 69.15187495290228,
     "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": 1.1173389949981356,
-      "bytes": 12086613,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 0.9005163840010937,
       "bytes": 1099069,
@@ -562,6 +556,12 @@
       "bytes": 1225280,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.2510575380001683,
+      "bytes": 12042377,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -273,3 +273,4 @@
 {"step": 512300, "loss": 0.5672863721847534, "lr": 0.0003, "grad_norm": 0.17085796645715126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74983471338, "tokens_per_second": 152572.5015712951, "data_wait_s": 1.5751000319141895e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 512400, "loss": 0.5869703888893127, "lr": 0.0003, "grad_norm": 0.15479998636327552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74998130872, "tokens_per_second": 152838.09116581045, "data_wait_s": 9.548999514663592e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 512500, "loss": 0.639297366142273, "lr": 0.0003, "grad_norm": 0.20095579893718304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75012726662, "tokens_per_second": 149972.7389789655, "data_wait_s": 1.931599763338454e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 512600, "loss": 0.5850881338119507, "lr": 0.0003, "grad_norm": 0.1786587468873638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75027353912, "tokens_per_second": 151838.88302942016, "data_wait_s": 2.4209002731367946e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15b57cefd390a5c4c07102ae982f97fe0baf6ee1ae3a2fb993a32f434918c3a1
-size 586530
+oid sha256:c78e78d12a0ea985fb47780dc4328fda85b1188433deec8ac2a85f49a074ca16
+size 588648