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": 75802.62597831695,
-      "required_pages_per_second_upper_estimate": 0.006570379499801418
+      "required_token_positions_per_second": 75657.19374097067,
+      "required_pages_per_second_upper_estimate": 0.006557773802062878
     },
     {
       "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": 9475.328247289619,
-      "required_pages_per_second_upper_estimate": 0.0037798501066258253
+      "required_token_positions_per_second": 9457.149217621334,
+      "required_pages_per_second_upper_estimate": 0.0037725982198904317
     },
     {
       "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": 18950.656494579238,
-      "required_pages_per_second_upper_estimate": 0.00438390085985773
+      "required_token_positions_per_second": 18914.29843524267,
+      "required_pages_per_second_upper_estimate": 0.004375490062710204
     },
     {
       "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": 28425.98474186886,
-      "required_pages_per_second_upper_estimate": 0.0017309933311639898
+      "required_token_positions_per_second": 28371.447652864,
+      "required_pages_per_second_upper_estimate": 0.0017276723085775862
     },
     {
       "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": 28425.98474186886,
-      "required_pages_per_second_upper_estimate": 0.0021883109935910967
+      "required_token_positions_per_second": 28371.447652864,
+      "required_pages_per_second_upper_estimate": 0.002184112577511178
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28425.98474186886,
-      "required_pages_per_second_upper_estimate": 0.3529555948429772
+      "required_token_positions_per_second": 28371.447652864,
+      "required_pages_per_second_upper_estimate": 0.3522784267214324
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.19801571489060166,
-    "producer_p95_seconds": 0.27088786700187484,
-    "gpu_update_seconds": 0.972631212289263,
-    "producer_batches_per_second": 5.050104233153783,
-    "gpu_batches_per_second": 1.02813891572153,
-    "producer_headroom": 4.911889002479502,
+    "producer_batch_seconds": 0.11466264789081038,
+    "producer_p95_seconds": 0.27059176500188187,
+    "gpu_update_seconds": 0.9745008551655286,
+    "producer_batches_per_second": 8.721235889757827,
+    "gpu_batches_per_second": 1.0261663647592594,
+    "producer_headroom": 8.498851832669303,
     "sustainable": true,
-    "buffer_seconds": 60.30313516193431,
+    "buffer_seconds": 60.41905302026277,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339879.5821235651,
-    "upload_bytes_per_second": 5876976.418564915,
+    "generation_bytes_per_second": 326733.271211824,
+    "upload_bytes_per_second": 5870206.444498992,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.93024633197989,
+    "estimated_full_shards_per_hour": 70.10935404077567,
     "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": 0.8282065210005385,
-      "bytes": 1197297,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 0.9917393740015541,
       "bytes": 7505816,
@@ -562,6 +556,12 @@
       "bytes": 6954584,
       "pages": 39,
       "shards": 2
+    },
+    {
+      "seconds": 1.6785005409983569,
+      "bytes": 5701009,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -178,3 +178,4 @@
 {"step": 132800, "loss": 0.6088363602757454, "lr": 0.0003, "grad_norm": 0.13583344653923674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19434292909, "tokens_per_second": 150707.80683408817, "data_wait_s": 1.588300074217841e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 132900, "loss": 0.5888306833803654, "lr": 0.0003, "grad_norm": 0.12400495761659745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19448992779, "tokens_per_second": 149753.48412678338, "data_wait_s": 1.1275998986093327e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133000, "loss": 0.6134922299534082, "lr": 0.0003, "grad_norm": 0.13954762211663865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19463662880, "tokens_per_second": 149523.63840803126, "data_wait_s": 1.3051998394075781e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 133100, "loss": 0.6408717334270477, "lr": 0.0003, "grad_norm": 0.15209618771773498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19478356154, "tokens_per_second": 150286.3180154748, "data_wait_s": 1.3798999134451151e-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 sha256:1c6350efa6c9843c4da794068911dbb9e6cb619a56e2761395ab734727b290b5
-size 383754
+oid sha256:0eb61eb6eb4deea73a203352f54a4fbffa457e7f04c8c10d3cca7577670f8210
+size 385872