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": 76892.66323212022,
-      "required_pages_per_second_upper_estimate": 0.005319612857957642
+      "required_token_positions_per_second": 76655.2729040576,
+      "required_pages_per_second_upper_estimate": 0.00530318964424082
     },
     {
       "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": 9611.582904015027,
-      "required_pages_per_second_upper_estimate": 0.004024169049236257
+      "required_token_positions_per_second": 9581.9091130072,
+      "required_pages_per_second_upper_estimate": 0.004011745252600499
     },
     {
       "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": 19223.165808030055,
-      "required_pages_per_second_upper_estimate": 0.004003678488839461
+      "required_token_positions_per_second": 19163.8182260144,
+      "required_pages_per_second_upper_estimate": 0.003991317952606586
     },
     {
       "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": 28834.748712045086,
-      "required_pages_per_second_upper_estimate": 0.00172859915717513
+      "required_token_positions_per_second": 28745.727339021596,
+      "required_pages_per_second_upper_estimate": 0.001723262461790138
     },
     {
       "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": 28834.748712045086,
-      "required_pages_per_second_upper_estimate": 0.002241774881705798
+      "required_token_positions_per_second": 28745.727339021596,
+      "required_pages_per_second_upper_estimate": 0.0022348538615168598
     },
     {
       "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": 28834.748712045086,
-      "required_pages_per_second_upper_estimate": 0.3698611961371081
+      "required_token_positions_per_second": 28745.727339021596,
+      "required_pages_per_second_upper_estimate": 0.36871932554766607
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10637939447650524,
-    "producer_p95_seconds": 0.303156702000706,
-    "gpu_update_seconds": 0.9588431054525075,
-    "producer_batches_per_second": 9.400316714726724,
-    "gpu_batches_per_second": 1.0429234921891306,
-    "producer_headroom": 9.013428870985686,
+    "producer_batch_seconds": 0.7250752405937533,
+    "producer_p95_seconds": 0.31156793300033314,
+    "gpu_update_seconds": 0.9618125043045455,
+    "producer_batches_per_second": 1.3791672146757004,
+    "gpu_batches_per_second": 1.039703679796788,
+    "producer_headroom": 1.32650027260196,
     "sustainable": true,
-    "buffer_seconds": 60.40711564350797,
+    "buffer_seconds": 60.59418777118636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426885.8312425613,
-    "upload_bytes_per_second": 4897418.970311788,
+    "generation_bytes_per_second": 337268.92567774217,
+    "upload_bytes_per_second": 4858364.453914596,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.59976199109677,
+    "estimated_full_shards_per_hour": 72.370060231678,
     "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": 1.1650056679973204,
-      "bytes": 7481151,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.9575125699993805,
       "bytes": 1008516,
@@ -562,6 +556,12 @@
       "bytes": 923574,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.519131245004246,
+      "bytes": 5792384,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 640000, "loss": 0.5932347774505615, "lr": 0.0001, "grad_norm": 0.19791672589781517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93679740292, "tokens_per_second": 150821.95741836695, "data_wait_s": 1.6641002730466425e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 640100, "loss": 0.5298950672149658, "lr": 0.0001, "grad_norm": 0.1671258591212648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93694299086, "tokens_per_second": 134443.71677900018, "data_wait_s": 1.1144999007228762e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 640200, "loss": 0.5764673352241516, "lr": 0.0001, "grad_norm": 0.1832202280649628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93708883486, "tokens_per_second": 151786.39544143138, "data_wait_s": 1.0358999134041369e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 640300, "loss": 0.5280592441558838, "lr": 0.0001, "grad_norm": 0.1931041105917027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93723538298, "tokens_per_second": 142031.75170983784, "data_wait_s": 1.463800435885787e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:224f10caabb77ece1e6af2ee3926c5ca9ec8667040f462852724ba2aa4500cb4
-size 764673
+oid sha256:abc71aa5fbbcdc520301b42a31102c6e4bd8b1f6965a90000ba7ec48e088dfb1
+size 766791