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": 75363.60475573018,
-      "required_pages_per_second_upper_estimate": 0.005213829044656508
+      "required_token_positions_per_second": 75608.23823963403,
+      "required_pages_per_second_upper_estimate": 0.005230753356700863
     },
     {
       "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": 9420.450594466272,
-      "required_pages_per_second_upper_estimate": 0.003944145942524682
+      "required_token_positions_per_second": 9451.029779954253,
+      "required_pages_per_second_upper_estimate": 0.003956948808922493
     },
     {
       "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": 18840.901188932545,
-      "required_pages_per_second_upper_estimate": 0.003924062849677373
+      "required_token_positions_per_second": 18902.059559908506,
+      "required_pages_per_second_upper_estimate": 0.003936800525496963
     },
     {
       "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": 28261.35178339882,
-      "required_pages_per_second_upper_estimate": 0.0016942248868292017
+      "required_token_positions_per_second": 28353.089339862763,
+      "required_pages_per_second_upper_estimate": 0.0016997244132640772
     },
     {
       "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": 28261.35178339882,
-      "required_pages_per_second_upper_estimate": 0.0021971957926100955
+      "required_token_positions_per_second": 28353.089339862763,
+      "required_pages_per_second_upper_estimate": 0.0022043279841143008
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28261.35178339882,
-      "required_pages_per_second_upper_estimate": 0.36250627600208846
+      "required_token_positions_per_second": 28353.089339862763,
+      "required_pages_per_second_upper_estimate": 0.3636829868762941
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09114903573444622,
-    "producer_p95_seconds": 0.27570068600107334,
-    "gpu_update_seconds": 0.9782971533669133,
-    "producer_batches_per_second": 10.971043104760888,
-    "gpu_batches_per_second": 1.0221843092953855,
-    "producer_headroom": 10.73294023885328,
+    "producer_batch_seconds": 0.2876716992111028,
+    "producer_p95_seconds": 0.2831828000016685,
+    "gpu_update_seconds": 0.9751318337338483,
+    "producer_batches_per_second": 3.4761848410613645,
+    "gpu_batches_per_second": 1.0255023632762863,
+    "producer_headroom": 3.389738498461974,
     "sustainable": true,
-    "buffer_seconds": 60.65442350874862,
+    "buffer_seconds": 60.458173691498594,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404107.96863877296,
+    "generation_bytes_per_second": 385407.6438210163,
     "upload_bytes_per_second": 5304296.640790743,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.71216291782753,
+    "estimated_full_shards_per_hour": 82.69950853321902,
     "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": [

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 743900, "loss": 0.5258931517601013, "lr": 5e-05, "grad_norm": 0.2305391180916694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108886548112, "tokens_per_second": 150104.7790159702, "data_wait_s": 5.028300074627623e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744000, "loss": 0.5613822340965271, "lr": 5e-05, "grad_norm": 0.23755027711396634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108901249779, "tokens_per_second": 151497.57460553377, "data_wait_s": 1.0911000572377816e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744100, "loss": 0.5697230100631714, "lr": 5e-05, "grad_norm": 0.24556399169410378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108915801769, "tokens_per_second": 149881.75720913263, "data_wait_s": 1.5187997632892802e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 744200, "loss": 0.5490539073944092, "lr": 5e-05, "grad_norm": 0.22234718924079527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108930354542, "tokens_per_second": 149898.92008542002, "data_wait_s": 1.2029999197693542e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2810aa407fd6cf6476f37d3b4e6870d7344d9c319cf15b89a00991fc16f8b349
-size 513735
+oid sha256:a4957848766b4c68861ad909eae1b91143c25c00fb1336e35e68fc63897179fb
+size 515853