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": 75840.32303307486,
-      "required_pages_per_second_upper_estimate": 0.006573646984965521
+      "required_token_positions_per_second": 75876.21253129235,
+      "required_pages_per_second_upper_estimate": 0.006576757795709911
     },
     {
       "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": 9480.040379134358,
-      "required_pages_per_second_upper_estimate": 0.003781729846471341
+      "required_token_positions_per_second": 9484.526566411543,
+      "required_pages_per_second_upper_estimate": 0.0037835194536506873
     },
     {
       "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": 18960.080758268716,
-      "required_pages_per_second_upper_estimate": 0.004386080997400915
+      "required_token_positions_per_second": 18969.053132823086,
+      "required_pages_per_second_upper_estimate": 0.00438815659834567
     },
     {
       "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": 28440.121137403075,
-      "required_pages_per_second_upper_estimate": 0.0017318541634840881
+      "required_token_positions_per_second": 28453.579699234626,
+      "required_pages_per_second_upper_estimate": 0.0017326737192880127
     },
     {
       "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": 28440.121137403075,
-      "required_pages_per_second_upper_estimate": 0.0021893992524513683
+      "required_token_positions_per_second": 28453.579699234626,
+      "required_pages_per_second_upper_estimate": 0.0021904353297968445
     },
     {
       "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": 28440.121137403075,
-      "required_pages_per_second_upper_estimate": 0.3531311215640398
+      "required_token_positions_per_second": 28453.579699234626,
+      "required_pages_per_second_upper_estimate": 0.35329823185907877
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07819368633593626,
-    "producer_p95_seconds": 0.19779442800063407,
-    "gpu_update_seconds": 0.9721477579657242,
-    "producer_batches_per_second": 12.788756315999645,
-    "gpu_batches_per_second": 1.0286502147498218,
-    "producer_headroom": 12.432560779769048,
+    "producer_batch_seconds": 0.05443503662509386,
+    "producer_p95_seconds": 0.19334548699953302,
+    "gpu_update_seconds": 0.9716879314395617,
+    "producer_batches_per_second": 18.370521303902507,
+    "gpu_batches_per_second": 1.0291369972234747,
+    "producer_headroom": 17.850413845255428,
     "sustainable": true,
-    "buffer_seconds": 60.273160993874896,
+    "buffer_seconds": 60.244651749252824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 270274.02975649544,
-    "upload_bytes_per_second": 5077777.563666837,
+    "generation_bytes_per_second": 286327.08625250554,
+    "upload_bytes_per_second": 5015768.0447323825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.994515128337355,
+    "estimated_full_shards_per_hour": 61.43912735635161,
     "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": [
@@ -430,6 +430,12 @@
       "bytes": 13717263,
       "pages": 44,
       "shards": 4
+    },
+    {
+      "seconds": 0.882563871999082,
+      "bytes": 1071171,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -79,3 +79,4 @@
 {"step": 162900, "loss": 0.6072080656886101, "lr": 0.0003, "grad_norm": 0.153760050164428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23841652859, "tokens_per_second": 151678.14044052552, "data_wait_s": 1.113600046664942e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 163000, "loss": 0.6016788110136986, "lr": 0.0003, "grad_norm": 0.15832981353342057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23856216415, "tokens_per_second": 148949.74164682618, "data_wait_s": 2.034299905062653e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 163100, "loss": 0.6613490562886, "lr": 0.0003, "grad_norm": 0.15087534734498603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23870729047, "tokens_per_second": 148544.3063488557, "data_wait_s": 1.259300006495323e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 163200, "loss": 0.5926289483904839, "lr": 0.0003, "grad_norm": 0.14027349682384607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23885261115, "tokens_per_second": 148769.39593597362, "data_wait_s": 1.2353999409242533e-05, "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 sha256:0161a5a40f7e8f15de194dc332b67d9b5e3aa3282d60cbce43ebbb6791a4e6d5
-size 172506
+oid sha256:0d82ac485c5fa5a000963c39f980c03e293089d1aa55c0e00c2940a5aeaed78a
+size 174624