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": 77089.70197016903,
-      "required_pages_per_second_upper_estimate": 0.00533324445504872
+      "required_token_positions_per_second": 76567.6285831396,
+      "required_pages_per_second_upper_estimate": 0.00529712620675687
     },
     {
       "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": 9636.212746271129,
-      "required_pages_per_second_upper_estimate": 0.004034481049859294
+      "required_token_positions_per_second": 9570.95357289245,
+      "required_pages_per_second_upper_estimate": 0.0040071583967321465
     },
     {
       "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": 19272.425492542257,
-      "required_pages_per_second_upper_estimate": 0.004013937982058114
+      "required_token_positions_per_second": 19141.9071457849,
+      "required_pages_per_second_upper_estimate": 0.003986754452428824
     },
     {
       "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": 28908.638238813386,
-      "required_pages_per_second_upper_estimate": 0.001733028721482115
+      "required_token_positions_per_second": 28712.860718677348,
+      "required_pages_per_second_upper_estimate": 0.0017212921580849238
     },
     {
       "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": 28908.638238813386,
-      "required_pages_per_second_upper_estimate": 0.0022475194674063527
+      "required_token_positions_per_second": 28712.860718677348,
+      "required_pages_per_second_upper_estimate": 0.0022322986263500796
     },
     {
       "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": 28908.638238813386,
-      "required_pages_per_second_upper_estimate": 0.37080897165009924
+      "required_token_positions_per_second": 28712.860718677348,
+      "required_pages_per_second_upper_estimate": 0.3682977478313176
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07002523152340245,
-    "producer_p95_seconds": 0.1715504849998979,
-    "gpu_update_seconds": 0.956392334069862,
-    "producer_batches_per_second": 14.28056685061869,
-    "gpu_batches_per_second": 1.045596001114489,
-    "producer_headroom": 13.657824662103907,
+    "producer_batch_seconds": 0.4178735818281183,
+    "producer_p95_seconds": 0.19088114199985284,
+    "gpu_update_seconds": 0.9629134578713479,
+    "producer_batches_per_second": 2.393068247160273,
+    "gpu_batches_per_second": 1.038514927614198,
+    "producer_headroom": 2.304317620795224,
     "sustainable": true,
-    "buffer_seconds": 60.25271704640131,
+    "buffer_seconds": 60.66354784589492,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436202.7917167607,
-    "upload_bytes_per_second": 5839117.27441857,
+    "generation_bytes_per_second": 370307.3596068119,
+    "upload_bytes_per_second": 5925512.086321525,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.59896482112042,
+    "estimated_full_shards_per_hour": 79.45933905747668,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 1097352,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.40065100099946,
+      "bytes": 12237786,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -52,3 +52,4 @@
 {"step": 490200, "loss": 0.7364625930786133, "lr": 0.0003, "grad_norm": 0.2280283186482364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71748771623, "tokens_per_second": 154282.14708244952, "data_wait_s": 8.301999514515046e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 490300, "loss": 0.5763944983482361, "lr": 0.0003, "grad_norm": 0.18574894482687884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71763351892, "tokens_per_second": 152421.45844116813, "data_wait_s": 2.0497999685176183e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 490400, "loss": 0.5646252036094666, "lr": 0.0003, "grad_norm": 0.16441540950719388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71778078272, "tokens_per_second": 145941.402206718, "data_wait_s": 1.1252999684074894e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 490500, "loss": 0.7770769596099854, "lr": 0.0003, "grad_norm": 0.40147290756149695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71792656749, "tokens_per_second": 152495.53326282685, "data_wait_s": 1.71370002135518e-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 sha256:8edc9fb79aee0a13d6a135753ddbdd1e1a83ff1170774013c29dddd7379d26a2
-size 115320
+oid sha256:d6d070426bf738128dd890df6b1d9635562b3042bc63d822a8e7d2343e3053eb
+size 117438