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": 74813.40887502761,
-      "required_pages_per_second_upper_estimate": 0.005175765216999176
+      "required_token_positions_per_second": 76762.93118746138,
+      "required_pages_per_second_upper_estimate": 0.005310637694088328
     },
     {
       "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": 9351.676109378452,
-      "required_pages_per_second_upper_estimate": 0.003915351501792973
+      "required_token_positions_per_second": 9595.366398432672,
+      "required_pages_per_second_upper_estimate": 0.004017379536988069
     },
     {
       "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": 18703.352218756903,
-      "required_pages_per_second_upper_estimate": 0.003895415026600063
+      "required_token_positions_per_second": 19190.732796865344,
+      "required_pages_per_second_upper_estimate": 0.003996923547929874
     },
     {
       "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": 28055.028328135355,
-      "required_pages_per_second_upper_estimate": 0.001681856110723832
+      "required_token_positions_per_second": 28786.09919529802,
+      "required_pages_per_second_upper_estimate": 0.0017256826929296543
     },
     {
       "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": 28055.028328135355,
-      "required_pages_per_second_upper_estimate": 0.0021811550514843313
+      "required_token_positions_per_second": 28786.09919529802,
+      "required_pages_per_second_upper_estimate": 0.0022379925957653228
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28055.028328135355,
-      "required_pages_per_second_upper_estimate": 0.3598597802508351
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28786.09919529802,
+      "required_pages_per_second_upper_estimate": 0.36923717237205805
     }
   ],
   "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": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13664086575772671,
-    "producer_p95_seconds": 0.32618444499894395,
-    "gpu_update_seconds": 0.9854917869490383,
-    "producer_batches_per_second": 7.318454800872428,
-    "gpu_batches_per_second": 1.014721800062766,
-    "producer_headroom": 7.212277099417537,
+    "producer_batch_seconds": 0.10649591464840569,
+    "producer_p95_seconds": 0.32631171899993205,
+    "gpu_update_seconds": 0.9604635839133105,
+    "producer_batches_per_second": 9.39003156413541,
+    "gpu_batches_per_second": 1.041163888718823,
+    "producer_headroom": 9.018783369148606,
     "sustainable": true,
-    "buffer_seconds": 60.114999003891334,
+    "buffer_seconds": 60.509205786538566,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 531150.145187219,
-    "upload_bytes_per_second": 5702893.943641559,
+    "generation_bytes_per_second": 515238.13258959004,
+    "upload_bytes_per_second": 5673953.04677683,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.97245661461285,
+    "estimated_full_shards_per_hour": 110.55810912385726,
     "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": 1.1392915559990797,
-      "bytes": 8097285,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.2682354990029125,
       "bytes": 6131054,
@@ -562,6 +556,12 @@
       "bytes": 7789873,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.150511206000374,
+      "bytes": 5980828,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -316,3 +316,4 @@
 {"step": 588700, "loss": 0.5562505125999451, "lr": 0.0003, "grad_norm": 0.15879911997357984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86167928257, "tokens_per_second": 149965.56399318564, "data_wait_s": 1.1205000191694126e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 588800, "loss": 0.5910505056381226, "lr": 0.0003, "grad_norm": 0.17230694447526343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86182519621, "tokens_per_second": 151182.607783369, "data_wait_s": 8.678998710820451e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 588900, "loss": 0.5647330284118652, "lr": 0.0003, "grad_norm": 0.1624249155753471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86197233219, "tokens_per_second": 152589.67730645378, "data_wait_s": 1.138900188379921e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 589000, "loss": 0.5505220890045166, "lr": 0.0003, "grad_norm": 0.15693049434588788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86211931171, "tokens_per_second": 152834.818940623, "data_wait_s": 1.4011999155627564e-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 sha256:8373d63c7ad4686b5504378469007af673a89f51796e39bc1b223f20b6230aa1
-size 678387
+oid sha256:36e680b974192b776aee10cc8a309738c3b04686418d2e424ab1fa2ac043d149
+size 680505