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": 77099.29732075719,
-      "required_pages_per_second_upper_estimate": 0.00533390828366669
+      "required_token_positions_per_second": 77032.99193365454,
+      "required_pages_per_second_upper_estimate": 0.0053293211231372345
     },
     {
       "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": 9637.412165094649,
-      "required_pages_per_second_upper_estimate": 0.004034983221473989
+      "required_token_positions_per_second": 9629.123991706818,
+      "required_pages_per_second_upper_estimate": 0.004031513136353245
     },
     {
       "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": 19274.824330189298,
-      "required_pages_per_second_upper_estimate": 0.004014437596678389
+      "required_token_positions_per_second": 19258.247983413636,
+      "required_pages_per_second_upper_estimate": 0.00401098518079268
     },
     {
       "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": 28912.236495283945,
-      "required_pages_per_second_upper_estimate": 0.0017332444314633053
+      "required_token_positions_per_second": 28887.37197512045,
+      "required_pages_per_second_upper_estimate": 0.001731753841445428
     },
     {
       "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": 28912.236495283945,
-      "required_pages_per_second_upper_estimate": 0.002247799216019881
+      "required_token_positions_per_second": 28887.37197512045,
+      "required_pages_per_second_upper_estimate": 0.0022458661089446354
     },
     {
       "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": 28912.236495283945,
-      "required_pages_per_second_upper_estimate": 0.3708551262205968
+      "required_token_positions_per_second": 28887.37197512045,
+      "required_pages_per_second_upper_estimate": 0.3705361908533812
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07499458503126277,
-    "producer_p95_seconds": 0.17569339699912234,
-    "gpu_update_seconds": 0.9562733067886269,
-    "producer_batches_per_second": 13.334296063950923,
-    "gpu_batches_per_second": 1.045726146386138,
-    "producer_headroom": 12.75123139077292,
+    "producer_batch_seconds": 0.09176198455476481,
+    "producer_p95_seconds": 0.18087010999988706,
+    "gpu_update_seconds": 0.9570964095942035,
+    "producer_batches_per_second": 10.897759075853315,
+    "gpu_batches_per_second": 1.0448268220167989,
+    "producer_headroom": 10.430206084121854,
     "sustainable": true,
-    "buffer_seconds": 60.245218327683496,
+    "buffer_seconds": 60.297073804434824,
     "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": 255255.8439189479,
-    "upload_bytes_per_second": 4580021.477777073,
+    "generation_bytes_per_second": 255920.60346810485,
+    "upload_bytes_per_second": 4571425.307156558,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.771962053073196,
+    "estimated_full_shards_per_hour": 54.914603977511966,
     "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": 0.8446817050003119,
-      "bytes": 2246420,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.1746671439996135,
       "bytes": 6226498,
@@ -562,6 +556,12 @@
       "bytes": 11702648,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0858605960002023,
+      "bytes": 2526198,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -93,3 +93,4 @@
 {"step": 449500, "loss": 0.6102327704429626, "lr": 0.0003, "grad_norm": 0.14200678050190313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65792922550, "tokens_per_second": 153365.92923557613, "data_wait_s": 9.314999260823242e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 449600, "loss": 0.6116858720779419, "lr": 0.0003, "grad_norm": 0.15781677754937184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65807490947, "tokens_per_second": 122341.90125808115, "data_wait_s": 1.4140001439955086e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 449700, "loss": 0.5876931548118591, "lr": 0.0003, "grad_norm": 0.17943939720067453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65821948113, "tokens_per_second": 150426.37173868914, "data_wait_s": 1.3969000065117143e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 449800, "loss": 0.5955415368080139, "lr": 0.0003, "grad_norm": 0.1597044843710986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65836597883, "tokens_per_second": 152744.1483349692, "data_wait_s": 1.5227000403683633e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c81d026192049c7f50c99017a42b6b2244a791c65dcda298539676300197ff7
-size 202158
+oid sha256:668adf5423f4ec70483fab95920ae2c6e9c21634558ea7899858ca446cdab8bb
+size 204276