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": 74779.28805643032,
-      "required_pages_per_second_upper_estimate": 0.006481679162358535
+      "required_token_positions_per_second": 74081.41132388846,
+      "required_pages_per_second_upper_estimate": 0.006421188975934215
     },
     {
       "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": 9347.41100705379,
-      "required_pages_per_second_upper_estimate": 0.0037288220069625778
+      "required_token_positions_per_second": 9260.176415486058,
+      "required_pages_per_second_upper_estimate": 0.003694022824112836
     },
     {
       "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": 18694.82201410758,
-      "required_pages_per_second_upper_estimate": 0.004324718055333688
+      "required_token_positions_per_second": 18520.352830972115,
+      "required_pages_per_second_upper_estimate": 0.004284357680368048
     },
     {
       "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": 28042.233021161366,
-      "required_pages_per_second_upper_estimate": 0.001707624864762579
+      "required_token_positions_per_second": 27780.529246458176,
+      "required_pages_per_second_upper_estimate": 0.0016916884779367455
     },
     {
       "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": 28042.233021161366,
-      "required_pages_per_second_upper_estimate": 0.002158768723838281
+      "required_token_positions_per_second": 27780.529246458176,
+      "required_pages_per_second_upper_estimate": 0.002138622042819215
     },
     {
       "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": 28042.233021161366,
-      "required_pages_per_second_upper_estimate": 0.3481906828061806
+      "required_token_positions_per_second": 27780.529246458176,
+      "required_pages_per_second_upper_estimate": 0.3449411977905581
     }
   ],
   "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": 0.12040951782762477,
-    "producer_p95_seconds": 0.389029557001777,
-    "gpu_update_seconds": 0.9859414540609562,
-    "producer_batches_per_second": 8.30499131664637,
-    "gpu_batches_per_second": 1.014259006841774,
-    "producer_headroom": 8.188235214697938,
+    "producer_batch_seconds": 0.3227009694217031,
+    "producer_p95_seconds": 0.42656380900007207,
+    "gpu_update_seconds": 0.9952294196671911,
+    "producer_batches_per_second": 3.098844115008554,
+    "gpu_batches_per_second": 1.0047934478609004,
+    "producer_headroom": 3.0840608302190535,
     "sustainable": true,
-    "buffer_seconds": 60.14242869771833,
+    "buffer_seconds": 60.70899459969866,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441485.2679241031,
-    "upload_bytes_per_second": 4735096.780447158,
+    "generation_bytes_per_second": 402678.20668165607,
+    "upload_bytes_per_second": 4689764.353647351,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.73246124546355,
+    "estimated_full_shards_per_hour": 86.40536928498517,
     "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": 1.428139841002121,
-      "bytes": 6304931,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.987766936996195,
       "bytes": 2378061,
@@ -562,6 +556,12 @@
       "bytes": 1208624,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 3.448138318002748,
+      "bytes": 11782468,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -407,3 +407,4 @@
 {"step": 195700, "loss": 0.6065610535442829, "lr": 0.0003, "grad_norm": 0.15682711098481222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28642713141, "tokens_per_second": 150739.66410759767, "data_wait_s": 1.3383003533817828e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 195800, "loss": 0.6248058266937733, "lr": 0.0003, "grad_norm": 0.15313246632839422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28657488126, "tokens_per_second": 151671.82703529843, "data_wait_s": 6.287998985499144e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 195900, "loss": 0.596268106251955, "lr": 0.0003, "grad_norm": 0.13300833685127866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28672190723, "tokens_per_second": 148810.555096693, "data_wait_s": 7.950002327561378e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 196000, "loss": 0.5848307311534882, "lr": 0.0003, "grad_norm": 0.17119020365577028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28686768471, "tokens_per_second": 148656.61338342412, "data_wait_s": 1.133199839387089e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c918e6acb0bfa6a1307169672735926a343e1d77d7283e5b27f4d29975dc27c
-size 872691
+oid sha256:fa164ce564511f818a3ab4b2c34bf876dd189332a8c3dd8a425e622438959026
+size 874809