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": 77041.42165718443,
-      "required_pages_per_second_upper_estimate": 0.005329904310970686
+      "required_token_positions_per_second": 76718.50627184553,
+      "required_pages_per_second_upper_estimate": 0.0053075642753460175
     },
     {
       "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": 9630.177707148054,
-      "required_pages_per_second_upper_estimate": 0.0040319543050044105
+      "required_token_positions_per_second": 9589.813283980691,
+      "required_pages_per_second_upper_estimate": 0.0040150545639292415
     },
     {
       "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": 19260.355414296107,
-      "required_pages_per_second_upper_estimate": 0.004011424103068809
+      "required_token_positions_per_second": 19179.626567961383,
+      "required_pages_per_second_upper_estimate": 0.00399461041334013
     },
     {
       "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": 28890.53312144416,
-      "required_pages_per_second_upper_estimate": 0.0017319433473407433
+      "required_token_positions_per_second": 28769.439851942076,
+      "required_pages_per_second_upper_estimate": 0.0017246839907328099
     },
     {
       "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": 28890.53312144416,
-      "required_pages_per_second_upper_estimate": 0.0022461118741668906
+      "required_token_positions_per_second": 28769.439851942076,
+      "required_pages_per_second_upper_estimate": 0.0022366974050961057
     },
     {
       "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": 28890.53312144416,
-      "required_pages_per_second_upper_estimate": 0.3705767386442473
+      "required_token_positions_per_second": 28769.439851942076,
+      "required_pages_per_second_upper_estimate": 0.36902348420289727
     }
   ],
   "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": 0.12012242067990542,
-    "producer_p95_seconds": 0.3231012260002899,
-    "gpu_update_seconds": 0.9569916859539749,
-    "producer_batches_per_second": 8.324840561319824,
-    "gpu_batches_per_second": 1.0449411574596412,
-    "producer_headroom": 7.966803204075494,
+    "producer_batch_seconds": 0.21343695927336626,
+    "producer_p95_seconds": 0.32920794899837347,
+    "gpu_update_seconds": 0.9610197536791328,
+    "producer_batches_per_second": 4.685224168318561,
+    "gpu_batches_per_second": 1.0405613372374882,
+    "producer_headroom": 4.502592976169024,
     "sustainable": true,
-    "buffer_seconds": 60.29047621510042,
+    "buffer_seconds": 60.54424448178537,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308338.13761648576,
-    "upload_bytes_per_second": 4680928.332677589,
+    "generation_bytes_per_second": 332288.17366809846,
+    "upload_bytes_per_second": 4678456.850662254,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.16218658801012,
+    "estimated_full_shards_per_hour": 71.30130679638114,
     "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": 1.1982823690013902,
-      "bytes": 1172595,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.8298632039986842,
       "bytes": 11926092,
@@ -562,6 +556,12 @@
       "bytes": 6344982,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.540335444999073,
+      "bytes": 2545551,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -313,3 +313,4 @@
 {"step": 873800, "loss": 0.5340554118156433, "lr": 3e-05, "grad_norm": 0.30434055765120077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127899912309, "tokens_per_second": 151690.4765906996, "data_wait_s": 9.535000572213903e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 873900, "loss": 0.5336430668830872, "lr": 3e-05, "grad_norm": 0.31682278954037923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127914685429, "tokens_per_second": 152967.68867623864, "data_wait_s": 1.412599885952659e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 874000, "loss": 0.5015284419059753, "lr": 3e-05, "grad_norm": 0.32555418062398894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127929404004, "tokens_per_second": 153155.30740216115, "data_wait_s": 1.6776000848039985e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 874100, "loss": 0.49958536028862, "lr": 3e-05, "grad_norm": 0.3206786860098578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127943824634, "tokens_per_second": 149745.2708233356, "data_wait_s": 1.2484000762924552e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6266000c904f6653fe965e878ee2abd63dfd2ad3b4832b31133c4d6e02e7fda5
-size 672033
+oid sha256:a5930ebf856d43c713eb17011cbbfbf400930ab5ee8969aff222836f075f575d
+size 674151