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": 76801.12511776578,
-      "required_pages_per_second_upper_estimate": 0.0053132800388089145
+      "required_token_positions_per_second": 76921.8970485512,
+      "required_pages_per_second_upper_estimate": 0.0053216353212101365
     },
     {
       "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": 9600.140639720723,
-      "required_pages_per_second_upper_estimate": 0.004019378412117881
+      "required_token_positions_per_second": 9615.2371310689,
+      "required_pages_per_second_upper_estimate": 0.004025698997794775
     },
     {
       "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": 19200.281279441446,
-      "required_pages_per_second_upper_estimate": 0.003998912245040056
+      "required_token_positions_per_second": 19230.4742621378,
+      "required_pages_per_second_upper_estimate": 0.0040052006470932084
     },
     {
       "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": 28800.42191916217,
-      "required_pages_per_second_upper_estimate": 0.0017265413183557771
+      "required_token_positions_per_second": 28845.711393206697,
+      "required_pages_per_second_upper_estimate": 0.0017292563531717236
     },
     {
       "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": 28800.42191916217,
-      "required_pages_per_second_upper_estimate": 0.0022391061245467564
+      "required_token_positions_per_second": 28845.711393206697,
+      "required_pages_per_second_upper_estimate": 0.0022426271819463747
     },
     {
       "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": 28800.42191916217,
-      "required_pages_per_second_upper_estimate": 0.36942088889524466
+      "required_token_positions_per_second": 28845.711393206697,
+      "required_pages_per_second_upper_estimate": 0.3700018136402393
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3136934663283455,
-    "producer_p95_seconds": 0.34863216100347927,
-    "gpu_update_seconds": 0.9599859362339614,
-    "producer_batches_per_second": 3.187825400715525,
-    "gpu_batches_per_second": 1.0416819270530298,
-    "producer_headroom": 3.0602675518562963,
+    "producer_batch_seconds": 0.15749056890649626,
+    "producer_p95_seconds": 0.3766904349977267,
+    "gpu_update_seconds": 0.9584787014998436,
+    "producer_batches_per_second": 6.349586562187797,
+    "gpu_batches_per_second": 1.0433200012010524,
+    "producer_headroom": 6.0859434831866155,
     "sustainable": true,
-    "buffer_seconds": 60.47911398273957,
+    "buffer_seconds": 60.38415819449015,
     "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": 299669.8337823977,
-    "upload_bytes_per_second": 5375194.96942742,
+    "generation_bytes_per_second": 300533.30551603413,
+    "upload_bytes_per_second": 5344418.309267631,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.3021703730006,
+    "estimated_full_shards_per_hour": 64.48745130644576,
     "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": 1.3053653089955333,
-      "bytes": 10336652,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.1407873469943297,
       "bytes": 7506775,
@@ -562,6 +556,12 @@
       "bytes": 10548156,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1907792190031614,
+      "bytes": 7251298,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 481000, "loss": 0.6311573386192322, "lr": 0.0003, "grad_norm": 0.19809674490948792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70402954995, "tokens_per_second": 150483.00139091918, "data_wait_s": 1.1340001947246492e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 481100, "loss": 0.5837187170982361, "lr": 0.0003, "grad_norm": 0.1556957615792114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70417428103, "tokens_per_second": 135862.18872660116, "data_wait_s": 1.6240999684669077e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 481200, "loss": 0.5885299444198608, "lr": 0.0003, "grad_norm": 0.16211804395881238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70432081063, "tokens_per_second": 150732.07105640034, "data_wait_s": 1.002899807645008e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 481300, "loss": 0.5504758358001709, "lr": 0.0003, "grad_norm": 0.13788691205636375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70446602724, "tokens_per_second": 150452.73208961496, "data_wait_s": 1.371200050925836e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b9067a5085b0d8a38b6e7e288d39a7e42a5f6401d3eea37adb469c04229333c
-size 874809
+oid sha256:d1cf5bed36acf018302842214233213b23e7bcd7fba0b726fe06e0d03b6ecb95
+size 876927