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": 76530.86688773091,
-      "required_pages_per_second_upper_estimate": 0.005294582947369612
+      "required_token_positions_per_second": 76497.70622166511,
+      "required_pages_per_second_upper_estimate": 0.005292288815547835
     },
     {
       "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": 9566.358360966364,
-      "required_pages_per_second_upper_estimate": 0.0040052344774576315
+      "required_token_positions_per_second": 9562.213277708139,
+      "required_pages_per_second_upper_estimate": 0.004003499017656594
     },
     {
       "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": 19132.716721932728,
-      "required_pages_per_second_upper_estimate": 0.003984840329508192
+      "required_token_positions_per_second": 19124.426555416278,
+      "required_pages_per_second_upper_estimate": 0.003983113706449209
     },
     {
       "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": 28699.07508289909,
-      "required_pages_per_second_upper_estimate": 0.001720465730269464
+      "required_token_positions_per_second": 28686.639833124413,
+      "required_pages_per_second_upper_estimate": 0.0017197202560329994
     },
     {
       "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": 28699.07508289909,
-      "required_pages_per_second_upper_estimate": 0.0022312268538049232
+      "required_token_positions_per_second": 28686.639833124413,
+      "required_pages_per_second_upper_estimate": 0.0022302600678318256
     },
     {
       "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": 28699.07508289909,
-      "required_pages_per_second_upper_estimate": 0.3681209204974165
+      "required_token_positions_per_second": 28686.639833124413,
+      "required_pages_per_second_upper_estimate": 0.3679614144652379
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0896557654219805,
-    "producer_p95_seconds": 0.25082212800043635,
-    "gpu_update_seconds": 0.9633759945272454,
-    "producer_batches_per_second": 11.153772379203117,
-    "gpu_batches_per_second": 1.0380163152090238,
-    "producer_headroom": 10.745276558545322,
+    "producer_batch_seconds": 0.097009286460775,
+    "producer_p95_seconds": 0.24951330299882102,
+    "gpu_update_seconds": 0.9637936042992007,
+    "producer_batches_per_second": 10.308291468614634,
+    "gpu_batches_per_second": 1.037566544890206,
+    "producer_headroom": 9.935065388702798,
     "sustainable": true,
-    "buffer_seconds": 60.69268765521646,
+    "buffer_seconds": 60.718997070849646,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315480.9268668324,
-    "upload_bytes_per_second": 5552280.816436256,
+    "generation_bytes_per_second": 303153.7820880952,
+    "upload_bytes_per_second": 5505069.092413298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.69486288551073,
+    "estimated_full_shards_per_hour": 65.04974457723753,
     "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": 1.4827766799971869,
-      "bytes": 15946512,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.33145204000175,
       "bytes": 7515329,
@@ -562,6 +556,12 @@
       "bytes": 5919198,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.178343194998888,
+      "bytes": 10619543,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -256,3 +256,4 @@
 {"step": 582700, "loss": 0.5577417612075806, "lr": 0.0003, "grad_norm": 0.17495125828870997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85290500663, "tokens_per_second": 150200.08823705386, "data_wait_s": 1.86079996638e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 582800, "loss": 0.6232706308364868, "lr": 0.0003, "grad_norm": 0.17642637879539783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85305100031, "tokens_per_second": 152282.8505410679, "data_wait_s": 1.42680000863038e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 582900, "loss": 0.5519506335258484, "lr": 0.0003, "grad_norm": 0.14989964877842898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85319728459, "tokens_per_second": 152525.82074454072, "data_wait_s": 1.2092001270502806e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 583000, "loss": 0.553880512714386, "lr": 0.0003, "grad_norm": 0.16887774106607287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85334310790, "tokens_per_second": 151980.73036090282, "data_wait_s": 1.0558000212768093e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f0d708a1d11844a58cb06ae3d8859c770de2cc3fce7c7a96600e13a634100a1
-size 550524
+oid sha256:712a04129e27ea87dcd62307a658443e09706668a6cafcb25195b9a984c54f89
+size 552642