Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75040.98366692908,
-      "required_pages_per_second_upper_estimate": 0.006504362274615107
+      "required_token_positions_per_second": 75123.51962072129,
+      "required_pages_per_second_upper_estimate": 0.006511516281904353
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9380.122958366135,
-      "required_pages_per_second_upper_estimate": 0.003741871293428329
+      "required_token_positions_per_second": 9390.43995259016,
+      "required_pages_per_second_upper_estimate": 0.0037459868966771026
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18760.24591673227,
-      "required_pages_per_second_upper_estimate": 0.00433985272378455
+      "required_token_positions_per_second": 18780.87990518032,
+      "required_pages_per_second_upper_estimate": 0.004344626033866213
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28140.368875098404,
-      "required_pages_per_second_upper_estimate": 0.001713600823388311
+      "required_token_positions_per_second": 28171.319857770482,
+      "required_pages_per_second_upper_estimate": 0.0017154855758457844
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28140.368875098404,
-      "required_pages_per_second_upper_estimate": 0.002166323493531785
+      "required_token_positions_per_second": 28171.319857770482,
+      "required_pages_per_second_upper_estimate": 0.0021687061858556853
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28140.368875098404,
-      "required_pages_per_second_upper_estimate": 0.3494092016725034
+      "required_token_positions_per_second": 28171.319857770482,
+      "required_pages_per_second_upper_estimate": 0.3497935092910151
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.1520139660702,
-    "producer_p95_seconds": 0.33098751500074286,
-    "gpu_update_seconds": 0.9825031122625366,
-    "producer_batches_per_second": 6.578342936847002,
-    "gpu_batches_per_second": 1.0178084807254921,
-    "producer_headroom": 6.463242408982454,
+    "producer_batch_seconds": 0.07994895105468913,
+    "producer_p95_seconds": 0.315188495998882,
+    "gpu_update_seconds": 0.981423665614086,
+    "producer_batches_per_second": 12.507981490788408,
+    "gpu_batches_per_second": 1.0189279462445922,
+    "producer_headroom": 12.275629044122699,
     "sustainable": true,
-    "buffer_seconds": 60.91519296027727,
+    "buffer_seconds": 60.84826726807333,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369237.1015773318,
+    "generation_bytes_per_second": 369229.982693017,
     "upload_bytes_per_second": 4689575.608156181,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.22968659868206,
+    "estimated_full_shards_per_hour": 79.22815905182726,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -252,3 +252,4 @@
 {"step": 262300, "loss": 0.625180721282959, "lr": 0.0003, "grad_norm": 0.15190336703338145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38393256750, "tokens_per_second": 148435.5978718008, "data_wait_s": 1.2223998055560514e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 262400, "loss": 0.5533402170985937, "lr": 0.0003, "grad_norm": 0.1547234429267112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38407988248, "tokens_per_second": 150214.3265424557, "data_wait_s": 1.4960001863073558e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 262500, "loss": 0.5813664942979813, "lr": 0.0003, "grad_norm": 0.145257338785955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38422674356, "tokens_per_second": 149571.74588261018, "data_wait_s": 1.6658999811625108e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 262600, "loss": 0.6075609959661961, "lr": 0.0003, "grad_norm": 0.13541707940987982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38437511449, "tokens_per_second": 151648.51463013145, "data_wait_s": 1.701099972706288e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:265a62016d594ff14b50d1a32ed3b4dac869175ddd4d496e67e244678261be23
-size 542052
+oid sha256:0989f1e53dcbe61753802effda091a8a633871236e14a6d94c940e12f336fa2b
+size 544170