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": 76754.5604656177,
-      "required_pages_per_second_upper_estimate": 0.005310058588128432
+      "required_token_positions_per_second": 76500.85398979121,
+      "required_pages_per_second_upper_estimate": 0.005292506585450627
     },
     {
       "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": 9594.320058202213,
-      "required_pages_per_second_upper_estimate": 0.004016941456183645
+      "required_token_positions_per_second": 9562.606748723902,
+      "required_pages_per_second_upper_estimate": 0.004003663755754285
     },
     {
       "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": 19188.640116404425,
-      "required_pages_per_second_upper_estimate": 0.003996487697777559
+      "required_token_positions_per_second": 19125.213497447803,
+      "required_pages_per_second_upper_estimate": 0.003983277605721319
     },
     {
       "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": 28782.96017460664,
-      "required_pages_per_second_upper_estimate": 0.0017254945134322131
+      "required_token_positions_per_second": 28687.820246171705,
+      "required_pages_per_second_upper_estimate": 0.001719791019992798
     },
     {
       "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": 28782.96017460664,
-      "required_pages_per_second_upper_estimate": 0.002237748550713661
+      "required_token_positions_per_second": 28687.820246171705,
+      "required_pages_per_second_upper_estimate": 0.002230351839754164
     },
     {
       "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": 28782.96017460664,
-      "required_pages_per_second_upper_estimate": 0.3691969083850469
+      "required_token_positions_per_second": 28687.820246171705,
+      "required_pages_per_second_upper_estimate": 0.3679765555363798
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08650056822659735,
-    "producer_p95_seconds": 0.28503144400019664,
-    "gpu_update_seconds": 0.9605683304385092,
-    "producer_batches_per_second": 11.560617698838632,
-    "gpu_batches_per_second": 1.0410503535375666,
-    "producer_headroom": 11.104763241811305,
+    "producer_batch_seconds": 0.1376842744452631,
+    "producer_p95_seconds": 0.28528834600001574,
+    "gpu_update_seconds": 0.9637539472414094,
+    "producer_batches_per_second": 7.262993570101237,
+    "gpu_batches_per_second": 1.0376092392278538,
+    "producer_headroom": 6.999738721974044,
     "sustainable": true,
-    "buffer_seconds": 60.515804817626076,
+    "buffer_seconds": 60.71649867620879,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 159,
+    "upload_pages_per_commit": 161,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 646706.6239216116,
-    "upload_bytes_per_second": 5195119.830838327,
+    "generation_bytes_per_second": 628376.1801194692,
+    "upload_bytes_per_second": 5182538.207553222,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.38409346693163,
+    "estimated_full_shards_per_hour": 67.4174502024081,
     "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": 1.1907369090004067,
-      "bytes": 6141214,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9464336839992029,
       "bytes": 2380354,
@@ -562,6 +556,12 @@
       "bytes": 11997860,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.7381740189994161,
+      "bytes": 7938212,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -321,3 +321,4 @@
 {"step": 752100, "loss": 0.5339847803115845, "lr": 5e-05, "grad_norm": 0.2317741747614533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110085950757, "tokens_per_second": 150050.41510498594, "data_wait_s": 9.28099689190276e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 752200, "loss": 0.5571590662002563, "lr": 5e-05, "grad_norm": 0.2566676285867222, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110100594628, "tokens_per_second": 151454.34595932436, "data_wait_s": 1.3125998520990834e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 752300, "loss": 0.5784069299697876, "lr": 5e-05, "grad_norm": 0.2570707362933527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110115272638, "tokens_per_second": 151708.78269091534, "data_wait_s": 1.6049001715146005e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 752400, "loss": 0.5917224287986755, "lr": 5e-05, "grad_norm": 0.29516761130177943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110130060104, "tokens_per_second": 152173.15408398607, "data_wait_s": 8.39999847812578e-06, "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 sha256:1549d5143db7510ab36ec8e3eddd51ad2dd39abb8808c66e26470df2e8b04191
-size 688977
+oid sha256:99f392663c086e81a29c218b75992e1d9f8ac6212ee19e77a2c1ad4fbd52ba16
+size 691095