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": 75280.79771540457,
-      "required_pages_per_second_upper_estimate": 0.006525148748533813
+      "required_token_positions_per_second": 75629.22489837605,
+      "required_pages_per_second_upper_estimate": 0.00655534953367316
     },
     {
       "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": 9410.09971442557,
-      "required_pages_per_second_upper_estimate": 0.0037538294696128813
+      "required_token_positions_per_second": 9453.653112297006,
+      "required_pages_per_second_upper_estimate": 0.0037712035712051247
     },
     {
       "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": 18820.19942885114,
-      "required_pages_per_second_upper_estimate": 0.004353721913667475
+      "required_token_positions_per_second": 18907.306224594013,
+      "required_pages_per_second_upper_estimate": 0.004373872537835342
     },
     {
       "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": 28230.299143276712,
-      "required_pages_per_second_upper_estimate": 0.0017190771048927176
+      "required_token_positions_per_second": 28360.959336891017,
+      "required_pages_per_second_upper_estimate": 0.0017270336251627727
     },
     {
       "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": 28230.299143276712,
-      "required_pages_per_second_upper_estimate": 0.0021732465745190666
+      "required_token_positions_per_second": 28360.959336891017,
+      "required_pages_per_second_upper_estimate": 0.0021833051579140607
     },
     {
       "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": 28230.299143276712,
-      "required_pages_per_second_upper_estimate": 0.35052583462603165
+      "required_token_positions_per_second": 28360.959336891017,
+      "required_pages_per_second_upper_estimate": 0.3521481969391834
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.41631964140646005,
-    "producer_p95_seconds": 0.4022427370000514,
-    "gpu_update_seconds": 0.9793732563611395,
-    "producer_batches_per_second": 2.40200053166284,
-    "gpu_batches_per_second": 1.021061166929858,
-    "producer_headroom": 2.352455082475824,
+    "producer_batch_seconds": 0.14163054111702422,
+    "producer_p95_seconds": 0.40006909200019436,
+    "gpu_update_seconds": 0.9748612404671508,
+    "producer_batches_per_second": 7.06062401592984,
+    "gpu_batches_per_second": 1.0257870130530604,
+    "producer_headroom": 6.883128686641521,
     "sustainable": true,
-    "buffer_seconds": 60.72114189439065,
+    "buffer_seconds": 60.441396908963355,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279105.0522586036,
-    "upload_bytes_per_second": 5790253.266318516,
+    "generation_bytes_per_second": 280694.9900838151,
+    "upload_bytes_per_second": 5859894.418154995,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.889446981607264,
+    "estimated_full_shards_per_hour": 60.23061062703934,
     "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": 1.197479630001908,
-      "bytes": 5919458,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8879908719973173,
       "bytes": 2645607,
@@ -562,6 +556,12 @@
       "bytes": 4760298,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.582079469000746,
+      "bytes": 13247311,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -391,3 +391,4 @@
 {"step": 234100, "loss": 0.6063950918614864, "lr": 0.0003, "grad_norm": 0.15298467270971142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34265917030, "tokens_per_second": 150056.52678457572, "data_wait_s": 1.572600012877956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 234200, "loss": 0.6237741932272911, "lr": 0.0003, "grad_norm": 0.1550310669100969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34280619420, "tokens_per_second": 136297.61284754198, "data_wait_s": 1.2418997357599437e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 234300, "loss": 0.5962309408932924, "lr": 0.0003, "grad_norm": 0.13739073586915035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34295097165, "tokens_per_second": 147068.08691130526, "data_wait_s": 8.19200067780912e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 234400, "loss": 0.6019500847905874, "lr": 0.0003, "grad_norm": 0.14625804607041817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34309682397, "tokens_per_second": 148176.22259378296, "data_wait_s": 9.702998795546591e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1906a21fa379fb147dcdd68559d0df8c7b180c3ed8a618914bc7c51d1b94670f
-size 838020
+oid sha256:16dfe39fc9436345587028f21c200594c46464d83542b14525bde64bc73df876
+size 840138