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": 77372.52574147326,
-      "required_pages_per_second_upper_estimate": 0.006706454432397234
+      "required_token_positions_per_second": 6802.549971698315,
+      "required_pages_per_second_upper_estimate": 0.0005896277906415307
     },
     {
       "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": 9671.565717684158,
-      "required_pages_per_second_upper_estimate": 0.0038581321675778516
+      "required_token_positions_per_second": 850.3187464622894,
+      "required_pages_per_second_upper_estimate": 0.0003392048613620111
     },
     {
       "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": 19343.131435368316,
-      "required_pages_per_second_upper_estimate": 0.00447469303008624
+      "required_token_positions_per_second": 1700.6374929245787,
+      "required_pages_per_second_upper_estimate": 0.000393412553790469
     },
     {
       "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": 29014.69715305247,
-      "required_pages_per_second_upper_estimate": 0.0017668428282697603
+      "required_token_positions_per_second": 2550.956239386868,
+      "required_pages_per_second_upper_estimate": 0.00015533985114562932
     },
     {
       "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": 29014.69715305247,
-      "required_pages_per_second_upper_estimate": 0.002233631704664268
+      "required_token_positions_per_second": 2550.956239386868,
+      "required_pages_per_second_upper_estimate": 0.00019637967280682777
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 29014.69715305247,
-      "required_pages_per_second_upper_estimate": 0.3602654326961828
+      "estimated_pages_for_buffer": 3,
+      "required_token_positions_per_second": 2550.956239386868,
+      "required_pages_per_second_upper_estimate": 0.03167433899185304
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,21 +155,21 @@
   "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": 63,
+    "prefetch_batches": 6,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 1.6148641634999876,
-    "producer_p95_seconds": 3.215140269000017,
-    "gpu_update_seconds": 0.9528963840000415,
-    "producer_batches_per_second": 0.6192471308748612,
-    "gpu_batches_per_second": 1.0494320440195484,
-    "producer_headroom": 0.5900783518130557,
-    "sustainable": false,
-    "buffer_seconds": 60.032472192002615,
-    "buffer_drain_seconds": 146.44865051046258,
+    "producer_batch_seconds": 0.053391263380955845,
+    "producer_p95_seconds": 0.1405949799999462,
+    "gpu_update_seconds": 10.838288628050044,
+    "producer_batches_per_second": 18.729656064978798,
+    "gpu_batches_per_second": 0.09226548898245328,
+    "producer_headroom": 202.99741833634823,
+    "sustainable": true,
+    "buffer_seconds": 65.02973176830027,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 6,
-    "max_pending_pages": 1460,
+    "upload_pages_per_commit": 17,
+    "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,
     "upload_sustainable": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -0,0 +1 @@
+{"step": 275100, "loss": 0.5665341280400753, "lr": 0.0003, "grad_norm": 0.15310209458591492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40266125057, "tokens_per_second": 49579.57407153421, "data_wait_s": 1.2700999945991498e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:00c467fbde6a65ef78e9234202fd330d3e101a8a5923e292e4b83a6710b110b1
-size 165
+oid sha256:4f62dbb9c6996bcd6b8b65ea08c1a0ad238a0f6b69ac6099dbddff5a0f6b5757
+size 2283