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": 75643.76148848394,
-      "required_pages_per_second_upper_estimate": 0.006556609528461061
+      "required_token_positions_per_second": 75700.32285632046,
+      "required_pages_per_second_upper_estimate": 0.006561512124471654
     },
     {
       "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": 9455.470186060493,
-      "required_pages_per_second_upper_estimate": 0.003771928429097053
+      "required_token_positions_per_second": 9462.540357040058,
+      "required_pages_per_second_upper_estimate": 0.003774748826009278
     },
     {
       "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": 18910.940372120986,
-      "required_pages_per_second_upper_estimate": 0.004374713233906895
+      "required_token_positions_per_second": 18925.080714080115,
+      "required_pages_per_second_upper_estimate": 0.004377984353157622
     },
     {
       "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": 28366.41055818148,
-      "required_pages_per_second_upper_estimate": 0.0017273655759390142
+      "required_token_positions_per_second": 28387.62107112017,
+      "required_pages_per_second_upper_estimate": 0.0017286571848940193
     },
     {
       "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": 28366.41055818148,
-      "required_pages_per_second_upper_estimate": 0.002183724808018947
+      "required_token_positions_per_second": 28387.62107112017,
+      "required_pages_per_second_upper_estimate": 0.002185357652019425
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28366.41055818148,
-      "required_pages_per_second_upper_estimate": 0.35221588286354694
+      "required_token_positions_per_second": 28387.62107112017,
+      "required_pages_per_second_upper_estimate": 0.3524792464472251
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.051446081546870914,
-    "producer_p95_seconds": 0.16661043899966899,
-    "gpu_update_seconds": 0.9746738997270039,
-    "producer_batches_per_second": 19.43782635979635,
-    "gpu_batches_per_second": 1.0259841781749668,
-    "producer_headroom": 18.94554202031906,
+    "producer_batch_seconds": 0.2644437009218734,
+    "producer_p95_seconds": 0.16535117300009006,
+    "gpu_update_seconds": 0.9739456480249902,
+    "producer_batches_per_second": 3.781523237323915,
+    "gpu_batches_per_second": 1.026751340824659,
+    "producer_headroom": 3.6829980998969996,
     "sustainable": true,
-    "buffer_seconds": 60.42978178307424,
+    "buffer_seconds": 60.38463017754939,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 566030.9718754684,
-    "upload_bytes_per_second": 4594557.390915708,
+    "generation_bytes_per_second": 516494.2735893789,
+    "upload_bytes_per_second": 4500469.795094228,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.72853501891154,
+    "estimated_full_shards_per_hour": 110.82764774094606,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 2574525,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 3.0401188460000412,
+      "bytes": 10860861,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 239700, "loss": 0.6767937075346708, "lr": 0.0003, "grad_norm": 0.14570195230763597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35084637670, "tokens_per_second": 147028.0926855148, "data_wait_s": 1.6646999938529916e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239800, "loss": 0.6235589403659105, "lr": 0.0003, "grad_norm": 0.15547576243761393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35099075885, "tokens_per_second": 146513.72239722084, "data_wait_s": 1.4210999779606936e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239900, "loss": 0.6314800642430782, "lr": 0.0003, "grad_norm": 0.16600692105706924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35113637319, "tokens_per_second": 148075.33592057668, "data_wait_s": 1.1266000001342036e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 240000, "loss": 0.5885987225919962, "lr": 0.0003, "grad_norm": 0.17048077004762824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35128325853, "tokens_per_second": 147986.62644008137, "data_wait_s": 1.0194999958912376e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47ac2532b6d6fd8602fbdbbc0d70f4ddfcb8bbf6e50292b2e1bdf1755f63a04f
-size 59469
+oid sha256:ed2048d0a61cebca21ce61b5876af95f188bb350a35ee1de9704ca0c11658d78
+size 61587