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": 74703.50796476098,
-      "required_pages_per_second_upper_estimate": 0.0064751107360755235
+      "required_token_positions_per_second": 74745.98036757372,
+      "required_pages_per_second_upper_estimate": 0.006478792136306011
     },
     {
       "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": 9337.938495595123,
-      "required_pages_per_second_upper_estimate": 0.003725043280515048
+      "required_token_positions_per_second": 9343.247545946715,
+      "required_pages_per_second_upper_estimate": 0.003727161140077675
     },
     {
       "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": 18675.876991190245,
-      "required_pages_per_second_upper_estimate": 0.004320335457702778
+      "required_token_positions_per_second": 18686.49509189343,
+      "required_pages_per_second_upper_estimate": 0.00432279176842827
     },
     {
       "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": 28013.815486785366,
-      "required_pages_per_second_upper_estimate": 0.0017058943860143608
+      "required_token_positions_per_second": 28029.742637840143,
+      "required_pages_per_second_upper_estimate": 0.0017068642659502932
     },
     {
       "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": 28013.815486785366,
-      "required_pages_per_second_upper_estimate": 0.002156581063377247
+      "required_token_positions_per_second": 28029.742637840143,
+      "required_pages_per_second_upper_estimate": 0.0021578071795546197
     },
     {
       "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": 28013.815486785366,
-      "required_pages_per_second_upper_estimate": 0.34783783213660013
+      "required_token_positions_per_second": 28029.742637840143,
+      "required_pages_per_second_upper_estimate": 0.3480355940479549
     }
   ],
   "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": 0.12078537472643802,
-    "producer_p95_seconds": 0.39892091599904234,
-    "gpu_update_seconds": 0.9869416043324077,
-    "producer_batches_per_second": 8.279148053022645,
-    "gpu_batches_per_second": 1.013231173567179,
-    "producer_headroom": 8.171035661955699,
+    "producer_batch_seconds": 0.09679603886723953,
+    "producer_p95_seconds": 0.3979328129935311,
+    "gpu_update_seconds": 0.9863808011806433,
+    "producer_batches_per_second": 10.331001265160742,
+    "gpu_batches_per_second": 1.013807242398732,
+    "producer_headroom": 10.190301304927493,
     "sustainable": true,
-    "buffer_seconds": 60.20343786427687,
+    "buffer_seconds": 60.16922887201924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 236238.00261351638,
-    "upload_bytes_per_second": 5899752.8934937315,
+    "generation_bytes_per_second": 255452.6980388508,
+    "upload_bytes_per_second": 5895673.3211902315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.691176021615206,
+    "estimated_full_shards_per_hour": 54.814202364674976,
     "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": 1.038481719999254,
-      "bytes": 2380066,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0937937710004917,
       "bytes": 10154198,
@@ -562,6 +556,12 @@
       "bytes": 7730044,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8456406199984485,
+      "bytes": 955055,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -385,3 +385,4 @@
 {"step": 153500, "loss": 0.6256589218974113, "lr": 0.0003, "grad_norm": 0.1370351269445419, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22464484527, "tokens_per_second": 148875.36384478706, "data_wait_s": 1.335500564891845e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 153600, "loss": 0.6513286847621202, "lr": 0.0003, "grad_norm": 0.13265619769425424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22479292121, "tokens_per_second": 151356.74003485168, "data_wait_s": 1.2650001735892147e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 153700, "loss": 0.6442008633166552, "lr": 0.0003, "grad_norm": 0.14107436594451314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22493950074, "tokens_per_second": 149837.5631263943, "data_wait_s": 7.187001756392419e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 153800, "loss": 0.5959294298663735, "lr": 0.0003, "grad_norm": 0.14721382194840682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22508767734, "tokens_per_second": 151025.75518803636, "data_wait_s": 1.1620999430306256e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6a3c7745d6d9936a0d880101ce4a6d90349af9dba33f035df1f975fbe63c94f6
-size 825312
+oid sha256:90b7d11961681fdd021ffca196742c64b366dc4a03327dcc9b9061425103ea78
+size 827430