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": 77107.71522175097,
-      "required_pages_per_second_upper_estimate": 0.0053344906535896615
+      "required_token_positions_per_second": 75353.350783516,
+      "required_pages_per_second_upper_estimate": 0.005213119651066238
     },
     {
       "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": 9638.464402718871,
-      "required_pages_per_second_upper_estimate": 0.00403542377139403
+      "required_token_positions_per_second": 9419.1688479395,
+      "required_pages_per_second_upper_estimate": 0.0039436093020198335
     },
     {
       "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": 19276.928805437743,
-      "required_pages_per_second_upper_estimate": 0.004014875903373896
+      "required_token_positions_per_second": 18838.337695879,
+      "required_pages_per_second_upper_estimate": 0.003923528941678177
     },
     {
       "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": 28915.393208156613,
-      "required_pages_per_second_upper_estimate": 0.0017334336715800003
+      "required_token_positions_per_second": 28257.5065438185,
+      "required_pages_per_second_upper_estimate": 0.0016939943705877026
     },
     {
       "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": 28915.393208156613,
-      "required_pages_per_second_upper_estimate": 0.0022480446365608186
+      "required_token_positions_per_second": 28257.5065438185,
+      "required_pages_per_second_upper_estimate": 0.0021968968421461474
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28915.393208156613,
-      "required_pages_per_second_upper_estimate": 0.37089561714391317
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28257.5065438185,
+      "required_pages_per_second_upper_estimate": 0.3624569533974487
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.40328230902341033,
-    "producer_p95_seconds": 0.21793054399677203,
-    "gpu_update_seconds": 0.956168909790267,
-    "producer_batches_per_second": 2.479652535271391,
-    "gpu_batches_per_second": 1.0458403214755718,
-    "producer_headroom": 2.3709666613091174,
+    "producer_batch_seconds": 0.41680167671864865,
+    "producer_p95_seconds": 0.22997905000011087,
+    "gpu_update_seconds": 0.9784302785925805,
+    "producer_batches_per_second": 2.399222593998883,
+    "gpu_batches_per_second": 1.0220452308962131,
+    "producer_headroom": 2.3474720310519404,
     "sustainable": true,
-    "buffer_seconds": 60.238641316786826,
+    "buffer_seconds": 60.66267727273999,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290697.8459376837,
-    "upload_bytes_per_second": 5108075.817065287,
+    "generation_bytes_per_second": 273184.20313328865,
+    "upload_bytes_per_second": 5031301.356945949,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.376990638712726,
+    "estimated_full_shards_per_hour": 58.61897058962817,
     "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": 1.142588306000107,
-      "bytes": 12693549,
-      "pages": 49,
-      "shards": 4
-    },
     {
       "seconds": 0.9721095380009501,
       "bytes": 837498,
@@ -562,6 +556,12 @@
       "bytes": 6648792,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.084016504999454,
+      "bytes": 5907243,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -196,3 +196,4 @@
 {"step": 459800, "loss": 0.6137932538986206, "lr": 0.0003, "grad_norm": 0.18723392024412436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67300855381, "tokens_per_second": 153362.58875885178, "data_wait_s": 1.6344998584827408e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 459900, "loss": 0.6186737418174744, "lr": 0.0003, "grad_norm": 0.1681104369439076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67315439425, "tokens_per_second": 150762.6402912328, "data_wait_s": 1.3208999007474631e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 460000, "loss": 0.599702775478363, "lr": 0.0003, "grad_norm": 0.16255731056963071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67329946830, "tokens_per_second": 150827.70684953185, "data_wait_s": 1.3354001566767693e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 460100, "loss": 0.549582302570343, "lr": 0.0003, "grad_norm": 0.1420083697173705, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67344454921, "tokens_per_second": 133765.46454767417, "data_wait_s": 1.1686999641824514e-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:b3f3bd11b3a20baefe21f6bfdac4bf0682c7c1cd23f4a03db47da1fc443f7e3f
-size 422661
+oid sha256:31d0cceb92badc7ea0fac1b16fc8ec90fba04bd94c162848d6ae94e25535f3e0
+size 424779