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": 74698.11404873568,
-      "required_pages_per_second_upper_estimate": 0.006474643205105205
+      "required_token_positions_per_second": 74537.06624910657,
+      "required_pages_per_second_upper_estimate": 0.006460683990005271
     },
     {
       "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": 9337.26425609196,
-      "required_pages_per_second_upper_estimate": 0.0037247743162964575
+      "required_token_positions_per_second": 9317.133281138322,
+      "required_pages_per_second_upper_estimate": 0.0037167437694025536
     },
     {
       "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": 18674.52851218392,
-      "required_pages_per_second_upper_estimate": 0.004320023510817084
+      "required_token_positions_per_second": 18634.266562276644,
+      "required_pages_per_second_upper_estimate": 0.004310709617292153
     },
     {
       "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": 28011.792768275875,
-      "required_pages_per_second_upper_estimate": 0.0017057712130695167
+      "required_token_positions_per_second": 27951.39984341496,
+      "required_pages_per_second_upper_estimate": 0.0017020936008026758
     },
     {
       "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": 28011.792768275875,
-      "required_pages_per_second_upper_estimate": 0.0021564253489071428
+      "required_token_positions_per_second": 27951.39984341496,
+      "required_pages_per_second_upper_estimate": 0.002151776134373034
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28011.792768275875,
-      "required_pages_per_second_upper_estimate": 0.3478127167423156
+      "required_token_positions_per_second": 27951.39984341496,
+      "required_pages_per_second_upper_estimate": 0.3470628387376605
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09958468945310983,
-    "producer_p95_seconds": 0.3250344070002029,
-    "gpu_update_seconds": 0.9870128709259951,
-    "producer_batches_per_second": 10.041704256866284,
-    "gpu_batches_per_second": 1.013158013898867,
-    "producer_headroom": 9.911291347559377,
+    "producer_batch_seconds": 0.10764798487497274,
+    "producer_p95_seconds": 0.2604311939976469,
+    "gpu_update_seconds": 0.9891454508498815,
+    "producer_batches_per_second": 9.289537571571316,
+    "gpu_batches_per_second": 1.0109736633179602,
+    "producer_headroom": 9.188703829418822,
     "sustainable": true,
-    "buffer_seconds": 60.2077851264857,
+    "buffer_seconds": 60.33787250184277,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279362.19075874955,
-    "upload_bytes_per_second": 5432853.28958258,
+    "generation_bytes_per_second": 294253.2109184397,
+    "upload_bytes_per_second": 5349317.899463305,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.94462291786065,
+    "estimated_full_shards_per_hour": 63.1398891989221,
     "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": 1.2312986870001623,
-      "bytes": 11373221,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0926308529997186,
       "bytes": 7644433,
@@ -562,6 +556,12 @@
       "bytes": 11939208,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.4417221109979437,
+      "bytes": 5916590,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -200,3 +200,4 @@
 {"step": 175000, "loss": 0.6425412036478519, "lr": 0.0003, "grad_norm": 0.15781174171328943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25612288862, "tokens_per_second": 150549.8516858283, "data_wait_s": 9.265000699087977e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 175100, "loss": 0.6106356605887413, "lr": 0.0003, "grad_norm": 0.13087914891331215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25626912939, "tokens_per_second": 129364.42735863585, "data_wait_s": 1.7846999980974942e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 175200, "loss": 0.6180292926728725, "lr": 0.0003, "grad_norm": 0.14005632310842198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25641595905, "tokens_per_second": 149659.5086331149, "data_wait_s": 1.4585999451810494e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 175300, "loss": 0.6606020256876945, "lr": 0.0003, "grad_norm": 0.16128645448699025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25656247799, "tokens_per_second": 149726.67388678485, "data_wait_s": 1.64319972100202e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6024a930594f7efe63aa4e287f1b5667548da958e8927cd708835dcc1c06d192
-size 431133
+oid sha256:02e294101e40fb72fdcf2ca9f6d9a2501767d9e378729ea44f04963b41108ece
+size 433251