Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77151.99303186582,
-      "required_pages_per_second_upper_estimate": 0.00533755389523208
+      "required_token_positions_per_second": 77294.53436202531,
+      "required_pages_per_second_upper_estimate": 0.005347415235193967
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9643.999128983227,
-      "required_pages_per_second_upper_estimate": 0.004037741045702689
+      "required_token_positions_per_second": 9661.816795253164,
+      "required_pages_per_second_upper_estimate": 0.0040452009304947295
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19287.998257966454,
-      "required_pages_per_second_upper_estimate": 0.004017181378414532
+      "required_token_positions_per_second": 19323.633590506328,
+      "required_pages_per_second_upper_estimate": 0.004024603278415672
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28931.997386949683,
-      "required_pages_per_second_upper_estimate": 0.0017344290667455318
+      "required_token_positions_per_second": 28985.450385759494,
+      "required_pages_per_second_upper_estimate": 0.0017376334924061767
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28931.997386949683,
-      "required_pages_per_second_upper_estimate": 0.002249335538427916
+      "required_token_positions_per_second": 28985.450385759494,
+      "required_pages_per_second_upper_estimate": 0.0022534912739704846
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28931.997386949683,
-      "required_pages_per_second_upper_estimate": 0.37110859772129245
+      "required_token_positions_per_second": 28985.450385759494,
+      "required_pages_per_second_upper_estimate": 0.3717942353966662
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05717903674995739,
-    "producer_p95_seconds": 0.19536767499994312,
-    "gpu_update_seconds": 0.9556201609665272,
-    "producer_batches_per_second": 17.488927006115492,
-    "gpu_batches_per_second": 1.0464408777108536,
-    "producer_headroom": 16.71277124071593,
+    "producer_batch_seconds": 0.2806389066561934,
+    "producer_p95_seconds": 0.24029365000023972,
+    "gpu_update_seconds": 0.9538578711746849,
+    "producer_batches_per_second": 3.5632978046949324,
+    "gpu_batches_per_second": 1.0483742182349354,
+    "producer_headroom": 3.3988796583477363,
     "sustainable": true,
-    "buffer_seconds": 60.20407014089122,
+    "buffer_seconds": 60.09304588400515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 417845.7831316817,
-    "upload_bytes_per_second": 3582531.9126001913,
+    "generation_bytes_per_second": 396604.184775878,
+    "upload_bytes_per_second": 3503544.3542698077,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.65997810805166,
+    "estimated_full_shards_per_hour": 85.10202557999854,
     "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": 1.6897458180001195,
-      "bytes": 13896820,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.1221129010000368,
       "bytes": 948347,
@@ -562,6 +556,12 @@
       "bytes": 5804474,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1589008739993005,
+      "bytes": 2402568,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 938700, "loss": 0.48941338062286377, "lr": 3e-05, "grad_norm": 0.34972946347485506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137399138972, "tokens_per_second": 153898.15498313733, "data_wait_s": 8.736999006941915e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 938800, "loss": 0.5446861386299133, "lr": 3e-05, "grad_norm": 0.35378563013557457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137413915310, "tokens_per_second": 153914.0110141069, "data_wait_s": 1.622700074221939e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 938900, "loss": 0.5586631894111633, "lr": 3e-05, "grad_norm": 0.35915213631929466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137428437368, "tokens_per_second": 151193.94303347502, "data_wait_s": 1.3079999916953966e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 939000, "loss": 0.5386180877685547, "lr": 3e-05, "grad_norm": 0.3467750174634002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137442993167, "tokens_per_second": 151637.46359498394, "data_wait_s": 9.938999937730841e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e24ab2a5a438679c29f6310acadbc951db7715e63267c0b5eea4371d0052d379
-size 296133
+oid sha256:993dbe223476072cd1739b0d580ddc36294f5f5f9b9cbe426c3c84188770831d
+size 298251