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": 76552.83571526268,
-      "required_pages_per_second_upper_estimate": 0.005296102801832964
+      "required_token_positions_per_second": 76031.43395664943,
+      "required_pages_per_second_upper_estimate": 0.005260031018353338
     },
     {
       "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": 9569.104464407836,
-      "required_pages_per_second_upper_estimate": 0.004006384213623415
+      "required_token_positions_per_second": 9503.929244581179,
+      "required_pages_per_second_upper_estimate": 0.003979096710095349
     },
     {
       "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": 19138.20892881567,
-      "required_pages_per_second_upper_estimate": 0.0039859842113626745
+      "required_token_positions_per_second": 19007.858489162358,
+      "required_pages_per_second_upper_estimate": 0.003958835652355149
     },
     {
       "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": 28707.313393223507,
-      "required_pages_per_second_upper_estimate": 0.0017209596043942415
+      "required_token_positions_per_second": 28511.78773374354,
+      "required_pages_per_second_upper_estimate": 0.0017092381396588128
     },
     {
       "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": 28707.313393223507,
-      "required_pages_per_second_upper_estimate": 0.0022318673461961492
+      "required_token_positions_per_second": 28511.78773374354,
+      "required_pages_per_second_upper_estimate": 0.0022166660861980673
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28707.313393223507,
-      "required_pages_per_second_upper_estimate": 0.3682265926966497
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28511.78773374354,
+      "required_pages_per_second_upper_estimate": 0.36571859947593716
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1524809701796812,
-    "producer_p95_seconds": 0.3411284679968958,
-    "gpu_update_seconds": 0.9630995287258903,
-    "producer_batches_per_second": 6.558195418232292,
-    "gpu_batches_per_second": 1.0383142865025863,
-    "producer_headroom": 6.316194916591813,
+    "producer_batch_seconds": 0.14777377254699786,
+    "producer_p95_seconds": 0.3256426799998735,
+    "gpu_update_seconds": 0.9697041889547582,
+    "producer_batches_per_second": 6.767100702406178,
+    "gpu_batches_per_second": 1.0312423225457008,
+    "producer_headroom": 6.562085898201957,
     "sustainable": true,
-    "buffer_seconds": 60.67527030973109,
+    "buffer_seconds": 60.12165971519501,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343030.3114896159,
-    "upload_bytes_per_second": 5395689.966192981,
+    "generation_bytes_per_second": 357007.79113373975,
+    "upload_bytes_per_second": 5354006.457211153,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.60631950870855,
+    "estimated_full_shards_per_hour": 76.60556126126427,
     "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": 1.7313189029991918,
-      "bytes": 10259217,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.5030511410004692,
       "bytes": 13228107,
@@ -562,6 +556,12 @@
       "bytes": 6045634,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.8855937030020868,
+      "bytes": 2335409,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 431800, "loss": 0.6038852334022522, "lr": 0.0003, "grad_norm": 0.16362444722250968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63203768086, "tokens_per_second": 152956.90786061552, "data_wait_s": 9.801005944609642e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 431900, "loss": 0.6109837293624878, "lr": 0.0003, "grad_norm": 0.15695645388663762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63218464732, "tokens_per_second": 151740.67150648794, "data_wait_s": 1.716199767543003e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 432000, "loss": 0.5201976895332336, "lr": 0.0003, "grad_norm": 0.14194316960438405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63233009864, "tokens_per_second": 149807.6279885452, "data_wait_s": 9.173993021249771e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 432100, "loss": 0.6040513515472412, "lr": 0.0003, "grad_norm": 0.1579544533545843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63247539191, "tokens_per_second": 150593.71209419233, "data_wait_s": 7.071997970342636e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d4daa097cea32fd50b893e6c705bbbdd667c94a901115b81aec912fcf895713
-size 744828
+oid sha256:90090b90b4e3fd79cbf4fdb4df953313102f13a6cde0807fb7d6e55ceae14175
+size 746946