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": 73514.50553416113,
-      "required_pages_per_second_upper_estimate": 0.0063720510188902384
+      "required_token_positions_per_second": 74293.61739552685,
+      "required_pages_per_second_upper_estimate": 0.006439582460392459
     },
     {
       "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": 9189.313191770141,
-      "required_pages_per_second_upper_estimate": 0.003665754424672946
+      "required_token_positions_per_second": 9286.702174440856,
+      "required_pages_per_second_upper_estimate": 0.0037046043459553436
     },
     {
       "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": 18378.626383540282,
-      "required_pages_per_second_upper_estimate": 0.004251571761054978
+      "required_token_positions_per_second": 18573.40434888171,
+      "required_pages_per_second_upper_estimate": 0.004296630215361608
     },
     {
       "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": 27567.93957531042,
-      "required_pages_per_second_upper_estimate": 0.0016787428823356512
+      "required_token_positions_per_second": 27860.106523322567,
+      "required_pages_per_second_upper_estimate": 0.0016965343165880821
     },
     {
       "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": 27567.93957531042,
-      "required_pages_per_second_upper_estimate": 0.002122256301448386
+      "required_token_positions_per_second": 27860.106523322567,
+      "required_pages_per_second_upper_estimate": 0.002144748121876237
     },
     {
       "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": 27567.93957531042,
-      "required_pages_per_second_upper_estimate": 0.3423015455667634
+      "required_token_positions_per_second": 27860.106523322567,
+      "required_pages_per_second_upper_estimate": 0.34592928124120054
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3369389500469424,
-    "producer_p95_seconds": 0.35387435499796993,
-    "gpu_update_seconds": 1.0029041134710437,
-    "producer_batches_per_second": 2.967896706096697,
-    "gpu_batches_per_second": 0.9971042959820031,
-    "producer_headroom": 2.9765158149015387,
+    "producer_batch_seconds": 0.10312454360951051,
+    "producer_p95_seconds": 0.3398366010005702,
+    "gpu_update_seconds": 0.9923867296363348,
+    "producer_batches_per_second": 9.69701261211474,
+    "gpu_batches_per_second": 1.0076716769141554,
+    "producer_headroom": 9.62318663337884,
     "sustainable": true,
-    "buffer_seconds": 60.17424680826262,
+    "buffer_seconds": 60.53559050781642,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 425634.45813842554,
+    "generation_bytes_per_second": 425626.83712799026,
     "upload_bytes_per_second": 5373224.897016624,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.33124645342421,
+    "estimated_full_shards_per_hour": 91.32961116199284,
     "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": [

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

@@ -254,3 +254,4 @@
 {"step": 180400, "loss": 0.5644677262753248, "lr": 0.0003, "grad_norm": 0.13632049778974314, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26403302958, "tokens_per_second": 151340.84074259398, "data_wait_s": 9.60000033956021e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 180500, "loss": 0.6260605715215206, "lr": 0.0003, "grad_norm": 0.17092064597473977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26417885651, "tokens_per_second": 148731.53539339133, "data_wait_s": 1.3316999684320763e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 180600, "loss": 0.6089816484600306, "lr": 0.0003, "grad_norm": 0.16993548851081852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26432370868, "tokens_per_second": 148228.87727112178, "data_wait_s": 1.5404999430757016e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 180700, "loss": 0.5505474973469973, "lr": 0.0003, "grad_norm": 0.1358629476055678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26446880042, "tokens_per_second": 147843.4271292146, "data_wait_s": 7.974998879944906e-06, "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 sha256:3ac2b38314b7280f4a3b1e2ed2674777c8ef1ae20e155eaff493b512689a93e5
-size 546288
+oid sha256:aaa1e8b827696c487745b12c419a5a1e16e2b45d83bd1a0875a61a2e8f58c6f4
+size 548406